Related
I need to draw a two-dimensional grid of Squares with centered Text on them onto a (transparent) PNG file.
The tiles need to have a sufficiently big resolution, so that the text does not get pixaleted to much.
For testing purposes I create a 2048x2048px 32-bit (transparency) PNG Image with 128x128px tiles like for example that one:
The problem is I need to do this with reasonable performance. All methods I have tried so far took more than 100ms to complete, while I would need this to be at a max < 10ms. Apart from that I would need the program generating these images to be Cross-Platform and support WebAssembly (but even if you have for example an idea how to do this using posix threads, etc. I would gladly take that as a starting point, too).
Net5 Implementation
using System.Diagnostics;
using System;
using System.Drawing;
namespace ImageGeneratorBenchmark
{
class Program
{
static int rowColCount = 16;
static int tileSize = 128;
static void Main(string[] args)
{
var watch = Stopwatch.StartNew();
Bitmap bitmap = new Bitmap(rowColCount * tileSize, rowColCount * tileSize);
Graphics graphics = Graphics.FromImage(bitmap);
Brush[] usedBrushes = { Brushes.Blue, Brushes.Red, Brushes.Green, Brushes.Orange, Brushes.Yellow };
int totalCount = rowColCount * rowColCount;
Random random = new Random();
StringFormat format = new StringFormat();
format.LineAlignment = StringAlignment.Center;
format.Alignment = StringAlignment.Center;
for (int i = 0; i < totalCount; i++)
{
int x = i % rowColCount * tileSize;
int y = i / rowColCount * tileSize;
graphics.FillRectangle(usedBrushes[random.Next(0, usedBrushes.Length)], x, y, tileSize, tileSize);
graphics.DrawString(i.ToString(), SystemFonts.DefaultFont, Brushes.Black, x + tileSize / 2, y + tileSize / 2, format);
}
bitmap.Save("Test.png");
watch.Stop();
Console.WriteLine($"Output took {watch.ElapsedMilliseconds} ms.");
}
}
}
This takes around 115ms on my machine. I am using the System.Drawing.Common nuget here.
Saving the bitmap takes roughly 55ms and drawing to the graphics object in the loop also takes roughly 60ms, while 40ms can be attributed to drawing the text.
Rust Implementation
use std::path::Path;
use std::time::Instant;
use image::{Rgba, RgbaImage};
use imageproc::{drawing::{draw_text_mut, draw_filled_rect_mut, text_size}, rect::Rect};
use rusttype::{Font, Scale};
use rand::Rng;
#[derive(Default)]
struct TextureAtlas {
segment_size: u16, // The side length of the tile
row_col_count: u8, // The amount of tiles in horizontal and vertical direction
current_segment: u32 // Points to the next segment, that will be used
}
fn main() {
let before = Instant::now();
let mut atlas = TextureAtlas {
segment_size: 128,
row_col_count: 16,
..Default::default()
};
let path = Path::new("test.png");
let colors = vec![Rgba([132u8, 132u8, 132u8, 255u8]), Rgba([132u8, 255u8, 32u8, 120u8]), Rgba([200u8, 255u8, 132u8, 255u8]), Rgba([255u8, 0u8, 0u8, 255u8])];
let mut image = RgbaImage::new(2048, 2048);
let font = Vec::from(include_bytes!("../assets/DejaVuSans.ttf") as &[u8]);
let font = Font::try_from_vec(font).unwrap();
let font_size = 40.0;
let scale = Scale {
x: font_size,
y: font_size,
};
// Draw random color rects for benchmarking
for i in 0..256 {
let rand_num = rand::thread_rng().gen_range(0..colors.len());
draw_filled_rect_mut(
&mut image,
Rect::at((atlas.current_segment as i32 % atlas.row_col_count as i32) * atlas.segment_size as i32, (atlas.current_segment as i32 / atlas.row_col_count as i32) * atlas.segment_size as i32)
.of_size(atlas.segment_size.into(), atlas.segment_size.into()),
colors[rand_num]);
let number = i.to_string();
//let text = &number[..];
let text = number.as_str(); // Somehow this conversion takes ~15ms here for 255 iterations, whereas it should normally only be less than 1us
let (w, h) = text_size(scale, &font, text);
draw_text_mut(
&mut image,
Rgba([0u8, 0u8, 0u8, 255u8]),
(atlas.current_segment % atlas.row_col_count as u32) * atlas.segment_size as u32 + atlas.segment_size as u32 / 2 - w as u32 / 2,
(atlas.current_segment / atlas.row_col_count as u32) * atlas.segment_size as u32 + atlas.segment_size as u32 / 2 - h as u32 / 2,
scale,
&font,
text);
atlas.current_segment += 1;
}
image.save(path).unwrap();
println!("Output took {:?}", before.elapsed());
}
For Rust I was using the imageproc crate. Previously I used the piet-common crate, but the output took more than 300ms. With the imageproc crate I got around 110ms in release mode, which is on par with the C# version, but I think it will perform better with webassembly.
When I used a static string instead of converting the number from the loop (see comment) I got below 100ms execution time. For Rust drawing to the image only takes around 30ms, but saving it takes 80ms.
C++ Implementation
#include <iostream>
#include <cstdlib>
#define cimg_display 0
#define cimg_use_png
#include "CImg.h"
#include <chrono>
#include <string>
using namespace cimg_library;
using namespace std;
/* Generate random numbers in an inclusive range. */
int random(int min, int max)
{
static bool first = true;
if (first)
{
srand(time(NULL));
first = false;
}
return min + rand() % ((max + 1) - min);
}
int main() {
auto t1 = std::chrono::high_resolution_clock::now();
static int tile_size = 128;
static int row_col_count = 16;
// Create 2048x2048px image.
CImg<unsigned char> image(tile_size*row_col_count, tile_size*row_col_count, 1, 3);
// Make some colours.
unsigned char cyan[] = { 0, 255, 255 };
unsigned char black[] = { 0, 0, 0 };
unsigned char yellow[] = { 255, 255, 0 };
unsigned char red[] = { 255, 0, 0 };
unsigned char green[] = { 0, 255, 0 };
unsigned char orange[] = { 255, 165, 0 };
unsigned char colors [] = { // This is terrible, but I don't now C++ very well.
cyan[0], cyan[1], cyan[2],
yellow[0], yellow[1], yellow[2],
red[0], red[1], red[2],
green[0], green[1], green[2],
orange[0], orange[1], orange[2],
};
int total_count = row_col_count * row_col_count;
for (size_t i = 0; i < total_count; i++)
{
int x = i % row_col_count * tile_size;
int y = i / row_col_count * tile_size;
int random_color_index = random(0, 4);
unsigned char current_color [] = { colors[random_color_index * 3], colors[random_color_index * 3 + 1], colors[random_color_index * 3 + 2] };
image.draw_rectangle(x, y, x + tile_size, y + tile_size, current_color, 1.0); // Force use of transparency. -> Does not work. Always outputs 24bit PNGs.
auto s = std::to_string(i);
CImg<unsigned char> imgtext;
unsigned char color = 1;
imgtext.draw_text(0, 0, s.c_str(), &color, 0, 1, 40); // Measure the text by drawing to an empty instance, so that the bounding box will be set automatically.
image.draw_text(x + tile_size / 2 - imgtext.width() / 2, y + tile_size / 2 - imgtext.height() / 2, s.c_str(), black, 0, 1, 40);
}
// Save result image as PNG (libpng and GraphicsMagick are required).
image.save_png("Test.png");
auto t2 = std::chrono::high_resolution_clock::now();
auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1).count();
std::cout << "Output took " << duration << "ms.";
getchar();
}
I also reimplemented the same program in C++ using CImg. For .png output libpng and GraphicsMagick are required, too. I am not very fluent in C++ and I did not even bother optimizing, because the save operation took ~200ms in Release mode, whereas the whole Image generation which is currently very unoptimized took only 30ms. So this solution also falls way short of my goal.
Where I am right now
A graph of where I am right now. I will update this when I make some progress.
Why I am trying to do this and why it bothers me so much
I was asked in the comments to give a bit more context. I know this question is getting a big bloated, but if you are interested read on...
So basically I need to build a Texture Atlas for a .gltf file. I need to generate a .gltf file from data and the primitives in the .gltf file will be assigned a texture based on the input data, too. In order to optimize for a small amount of draw calls I am putting as much geometry as possible into one single primitive and then use texture coordinates to map the texture to the model. Now GPUs have a maximum size, that the texture can have. I will use 2048x2048 pixels, because the majority of devices supports at least that. That means, that if I have more than 256 objects, I need to add a new primitive to the .gltf and generate another texture atlas. In some cases one texture atlas might be sufficient, in other cases I need up to 15-20.
The textures will have a (semi-)transparent background, maybe text and maybe some lines / hatches or simple symbols, that can be drawn with a path.
I have the whole system set up in Rust already and the .gltf generating is really efficient: I can generate 54000 vertecies (=1500 boxes for example) in about 10ms which is a common case. Now for this I need to generate 6 texture atlases, which is not really a problem on a multi-core system (7 threads one for the .gltf, six for the textures). The problem is generating one takes about 100ms (or now 55 ms) which makes the whole process more than 5 times slower.
Unfortunatly it gets even worse, because another common case is 15000 objects. Generating the vertecies (plus a lot of custom attributes actually) and assembling the .gltf still only takes 96ms (540000 Vertecies / 20MB .gltf), but in that time I need to generate 59 texture atlases. I am working on a 8-core System, so at that point it gets impossible for me to run them all in parallel and I will have to generate ~9 atlases per thread (which means 55ms*9 = 495ms) so again this is 5 times as much and actually creates a quite noticeable lag. In reality it currently takes more than 2.5 s, because I am have updated to use the faster code and there seems to be additional slowdown.
What I need to do
I do understand that it will take some time to write out 4194304 32-bit pixels. But as far as I can see, because I am only writing to different parts of the image (for example only to the upper tile and so on) it should be possible to build a program that does this using multiple threads. That is what I would like to try and I would take any hint on how to make my Rust program run faster.
If it helps I would also be willing to rewrite this in C or any other language, that can be compiled to wasm and can be called via Rust's FFI. So if you have suggestions for more performant libraries I would be very thankful for that too.
Edit
Update 1: I made all the suggested improvements for the C# version from the comments. Thanks for all of them. It is now at 115ms and almost exactly as fast as the Rust version, which makes me believe I am sort of hitting a dead-end there and I would really need to find a way to parallize this in order to make significant further improvements...
Update 2: Thanks to #pinkfloydx33 I was able to run the binary with around 60ms (including the first run) after publishing it with dotnet publish -p:PublishReadyToRun=true --runtime win10-x64 --configuration Release.
In the meantime I also tried other methods myself, namely Python with Pillow (~400ms), C# and Rust both with Skia (~314ms and ~260ms) and I also reimplemented the program in C++ using CImg (and libpng as well as GraphicsMagick).
I was able to get all of the drawing (creating the grid and the text) down to 4-5ms by:
Caching values where possible (Random, StringFormat, Math.Pow)
Using ArrayPool for scratch buffer
Using the DrawString overload accepting a StringFormat with the following options:
Alignment and LineAlignment for centering (in lieu of manually calculating)
FormatFlags and Trimming options that disable things like overflow/wrapping since we are just writing small numbers (this had an impact, though negligible)
Using a custom Font from the GenericMonospace font family instead of SystemFonts.DefaultFont
This shaved off ~15ms
Fiddling with various Graphics options, such as TextRenderingHint and SmoothingMode
I got varying results so you may want to fiddle some more
An array of Color and the ToArgb function to create an int representing the 4x bytes of the pixel's color
Using LockBits, (semi-)unsafe code and Span to
Fill a buffer representing 1px high and size * countpx wide (the entire image width) with the int representing the ARGB values of the random colors
Copy that buffer size times (now representing an entire square in height)
Rinse/Repeat
unsafe was required to create a Span<> from the locked bit's Scan0 pointer
Finally, using GDI/native to draw the text over the graphic
I was then able to shave a little bit of time off of the actual saving process by using the Image.Save(Stream) overload. I used a FileStream with a custom buffer-size of 16kb (over the default 4kb) which seemed to be the sweet spot. This brought the total end-to-end time down to around 40ms (on my machine).
private static readonly Random Random = new();
private static readonly Color[] UsedColors = { Color.Blue, Color.Red, Color.Green, Color.Orange, Color.Yellow };
private static readonly StringFormat Format = new()
{
Alignment = StringAlignment.Center,
LineAlignment = StringAlignment.Center,
FormatFlags = StringFormatFlags.NoWrap | StringFormatFlags.FitBlackBox | StringFormatFlags.NoClip,
Trimming = StringTrimming.None, HotkeyPrefix = HotkeyPrefix.None
};
private static unsafe void DrawGrid(int count, int size, bool save)
{
var intsPerRow = size * count;
var sizePerFullRow = intsPerRow * size;
var colorsLen = UsedColors.Length;
using var bitmap = new Bitmap(intsPerRow, intsPerRow, PixelFormat.Format32bppArgb);
var bmpData = bitmap.LockBits(new Rectangle(0, 0, bitmap.Width, bitmap.Height), ImageLockMode.WriteOnly, PixelFormat.Format32bppArgb);
var byteSpan = new Span<byte>(bmpData.Scan0.ToPointer(), Math.Abs(bmpData.Stride) * bmpData.Height);
var intSpan = MemoryMarshal.Cast<byte, int>(byteSpan);
var arr = ArrayPool<int>.Shared.Rent(intsPerRow);
var buff = arr.AsSpan(0, intsPerRow);
for (int y = 0, offset = 0; y < count; ++y)
{
// fill buffer with an entire 1px row of colors
for (var bOffset = 0; bOffset < intsPerRow; bOffset += size)
buff.Slice(bOffset, size).Fill(UsedColors[Random.Next(0, colorsLen)].ToArgb());
// duplicate the pixel high row until we've created a row of squares in full
var len = offset + sizePerFullRow;
for ( ; offset < len; offset += intsPerRow)
buff.CopyTo(intSpan.Slice(offset, intsPerRow));
}
ArrayPool<int>.Shared.Return(arr);
bitmap.UnlockBits(bmpData);
using var graphics = Graphics.FromImage(bitmap);
graphics.TextRenderingHint = TextRenderingHint.ClearTypeGridFit;
// some or all of these may not even matter?
// you may try removing/modifying the rest
graphics.CompositingQuality = CompositingQuality.HighSpeed;
graphics.InterpolationMode = InterpolationMode.Default;
graphics.SmoothingMode = SmoothingMode.HighSpeed;
graphics.PixelOffsetMode = PixelOffsetMode.HighSpeed;
var font = new Font(FontFamily.GenericMonospace, 14, FontStyle.Regular);
var lenSquares = count * count;
for (var i = 0; i < lenSquares; ++i)
{
var x = i % count * size;
var y = i / count * size;
var rect = new Rectangle(x, y, size, size);
graphics.DrawString(i.ToString(), font, Brushes.Black, rect, Format);
}
if (save)
{
using var fs = new FileStream("Test.png", FileMode.Create, FileAccess.Write, FileShare.Write, 16 * 1024);
bitmap.Save(fs, ImageFormat.Png);
}
}
Here are the timings (in ms) using a StopWatch in Release mode, run outside of Visual Studio. At least the first 1 or 2 timings should be ignored since the methods aren't fully jitted yet. Your mileage will vary depending on your PC, etc.
Image generation only:
Elapsed: 38
Elapsed: 6
Elapsed: 4
Elapsed: 4
Elapsed: 4
Elapsed: 4
Elapsed: 5
Elapsed: 4
Elapsed: 5
Elapsed: 4
Elapsed: 4
Image Generation and saving:
Elapsed: 95
Elapsed: 48
Elapsed: 41
Elapsed: 40
Elapsed: 37
Elapsed: 42
Elapsed: 42
Elapsed: 39
Elapsed: 38
Elapsed: 40
Elapsed: 41
I don't think there is anything that can be done about the slow save. I reviewed the source code of Image.Save. It calls into Native/GDI, passing in a Handle to the Stream, the native image pointer and the Guid representing PNG's ImageCodecInfo (encoder). Any slowness is going to be on that end. Update: I have verified that you get the same slow speed when saving to a MemoryStream so this has nothing to do with the fact you are saving to a file and everything to do with what's going on behind the scenes with GDI/native.
I also attempted to get the Image drawing down further using direct unsafe (pointers) and/or tricks with Unsafe and MemoryMarshal (ex. CopyBlock) as well as unrolling the loops. Those methods either produced identical results or worse and made things a bit harder to follow.
Note: Publishing as a console application with PublishReadyToRun=true seems to help a bit as well.
Update
I realize that the above is just an example, so this may not apply to your end goal. Upon further, extensive review I found that the bulk of the time spent is actually part of Image::Save. It doesn't matter what type of Stream we are saving to, even MemoryStream exhibits the same slowness (obviously disregarding file I/O). I am confident this is related to having GDI objects in the Image/Graphics--in our case the text from DrawString.
As a "simple" test I updated the above so that drawing of the text happened on a secondary image of all white. Without saving that image, I then looped over its individual pixels and based on the rough color (since we have aliasing to deal with) I manually set the corresponding pixel on the primary bitmap. The entire end to end process took sub 20ms on my machine. The rendered image wasn't perfect since it was a quick test, but it proves that you can do parts of this manually and still achieve really low times. The problem is the text drawing but we can leverage GDI without actually using it in our final image. You just need to find the sweet spot. I also tried using an indexed format and populating the pallette with colors beforehand also appeared to help some. Anyways, just food for thought.
I am trying to find if the image is clipped from the bottom and if it is, then I will divide it in two images from the last white pixel row. Following are the simple methods I created to check clipping and get the empty white pixel rows. Also, as you can see this is not a very good solution. This might cause performance issues for larger images. So if anyone can suggest me better ways then it will be a great help:
private static bool IsImageBottomClipping(Bitmap image)
{
for (int i = 0; i < image.Width; i++)
{
var pixel = image.GetPixel(i, image.Height - 1);
if (pixel.ToArgb() != Color.White.ToArgb())
{
return true;
}
}
return false;
}
private static int GetLastWhiteLine(Bitmap image)
{
for (int i = image.Height - 1; i >= 0; i--)
{
int whitePixels = 0;
for (int j = 0; j < image.Width; j++)
{
var pixel = image.GetPixel(j, i);
if (pixel.ToArgb() == Color.White.ToArgb())
{
whitePixels = j + 1;
}
}
if (whitePixels == image.Width)
return i;
}
return -1;
}
IsImageBottomClipping is working fine. But other method is not sending correct white pixel row. It is only sending one less row. Example image:
In this case, row around 180 should be the return value of GetLastWhiteLine method. But it is returning 192.
All right, so... we got two of subjects to tackle here. First, the optimising, then, your bug. I'll start with the optimising.
The fastest way is to work in memory directly, but, honestly, it's kind of unwieldy. The second-best choice, which is what I generally use, is to copy the raw image data bytes out of the image object. This will make you end up with four vital pieces of data:
The width, which you can just get from the image.
The height, which you can just get from the image.
The byte array, containing the image bytes.
The stride, which gives you the amount of bytes used for each line on the image.
(Technically, there's a fifth one, namely the pixel format, but we'll just force things to 32bpp here so we don't have to take that into account along the way.)
Note that the stride, technically, is not just the amount of bytes used per pixel multiplied by the image width. It is rounded up to the next multiple of 4 bytes. When working with 32-bit ARGB content, this isn't really an issue, since 32-bit is 4 bytes, but in general, it's better to use the stride and not just the multiplied width, and write all code assuming there could be padded bytes behind each line. You'll thank me if you're ever processing 24-bit RGB content with this kind of system.
However, when going over the image's content you obviously should only check the exact range that contains pixel data, and not the full stride.
The way to get these things is quite simple: use LockBits on the image, tell it to expose the image as 32 bit per pixel ARGB data (it will actually convert it if needed), get the line stride, and use Marshal.Copy to copy the entire image contents into a byte array.
Int32 width = image.Width;
Int32 height = image.Height;
BitmapData sourceData = image.LockBits(new Rectangle(0, 0, width, height), ImageLockMode.ReadOnly, PixelFormat.Format32bppArgb);
Int32 stride = sourceData.Stride;
Byte[] data = new Byte[stride * height];
Marshal.Copy(sourceData.Scan0, data, 0, data.Length);
image.UnlockBits(sourceData);
As mentioned, this is forced to 32-bit ARGB format. If you would want to use this system to get the data out in the original format it has inside the image, just change PixelFormat.Format32bppArgb to image.PixelFormat.
Now, you have to realise, LockBits is a rather heavy operation, which copies the data out, in the requested pixel format, to new memory, where it can be read or (if not specified as read-only as I did here) edited. What makes this more optimal than your method is, quite simply, that GetPixel performs a LockBits operation every time you request a single pixel value. So you're cutting down the amount of LockBits calls from several thousands to just one.
Anyway, now, as for your functions.
The first method is, in my opinion, completely unnecessary; you should just run the second one on any image you get. Its output gives you the last white line of the image, so if that value equals height-1 you're done, and if it doesn't, you immediately have the value needed for the further processing. The first function does exactly the same as the second, after all; it checks if all pixels on a line are white. The only difference is that it only processes the last line.
So, onto the second method. This is where things go wrong. You set the amount of white pixels to the "current pixel index plus one", rather than incrementing it to check if all pixels matched, meaning the method goes over all pixels but only really checks if the last pixel on the row was white. Since your image indeed has a white pixel at the end of the last row, it aborts after one row.
Also, whenever you find a pixel that does not match, you should just abort the scan of that line immediately, like your first method does; there's no point in continuing on that line after that.
So, let's fix that second function, and rewrite it to work with that set of "byte array", "stride", "width" and "height", rather than an image. I added the "white" colour as parameter too, to make it more reusable, so it's changed from GetLastWhiteLine to GetLastClearLine.
One general usability note: if you are iterating over the height and width, do actually call your loop variables y and x; it makes things a lot more clear in your code.
I explained the used systems in the code comments.
private static Int32 GetLastClearLine(Byte[] sourceData, Int32 stride, Int32 width, Int32 height, Color checkColor)
{
// Get color as UInt32 in advance.
UInt32 checkColVal = (UInt32)checkColor.ToArgb();
// Use MemoryStream with BinaryReader since it can read UInt32 from a byte array directly.
using (MemoryStream ms = new MemoryStream(sourceData))
using (BinaryReader sr = new BinaryReader(ms))
{
for (Int32 y = height - 1; y >= 0; --y)
{
// Set position in the memory stream to the start of the current row.
ms.Position = stride * y;
Int32 matchingPixels = 0;
// Read UInt32 pixels for the whole row length.
for (Int32 x = 0; x < width; ++x)
{
// Read a UInt32 for one whole 32bpp ARGB pixel.
UInt32 colorVal = sr.ReadUInt32();
// Compare with check value.
if (colorVal == checkColVal)
matchingPixels++;
else
break;
}
// Test if full line matched the given color.
if (matchingPixels == width)
return y;
}
}
return -1;
}
This can be simplified, though; the loop variable x already contains the value you need, so if you simply declare it before the loop, you can check after the loop what value it had when the loop stopped, and there is no need to increment a second variable. And, honestly, the value read from the stream can be compared directly, without the colorVal variable. Making the contents of the y-loop:
{
ms.Position = stride * y;
Int32 x;
for (x = 0; x < width; ++x)
if (sr.ReadUInt32() != checkColVal)
break;
if (x == width)
return y;
}
For your example image, this gets me value 178, which is correct when I check in Gimp.
I have a series of images like:
I want to remove all those small little irregular shapes, to get only big circular shape.
I have tried denoising:
Cv2.FastNlMeansDenoising(myMat, myMat,h:3);
But not getting nice results and process is slowed, so it appears it needs other processing, so I tried, dilate and blur:
int erosionSize = 2;
Mat element = Cv2.GetStructuringElement(MorphShapes.Cross,
new OpenCvSharp.Size(2 * erosionSize + 1, 2 * erosionSize + 1),
new OpenCvSharp.Point(erosionSize, erosionSize));
Cv2.Dilate(myMat, myMat, element, iterations: 2);
Cv2.Blur(myMat, myMat, new OpenCvSharp.Size(9, 9));
but getting something like
I guess maybe using hsv or something would help, what would be a better approach?
Say I have this 10x5 array:
..........
..........
..........
..........
..........
and this 1x2 array:
AB
CD
EF
Now I want to write the second array into the bigger one at position 1/2 (X-pos/Y-pos), deleting all old values (my example is zero based & inclusive). The result would be:
..........
..........
.AB.......
.CD.......
.EF.......
There might be multiple sub arrays with a known overwrite hierarchy, the arrays might have more than 3 dimensions and the arrays contain complex objects.
Is there a best practice to do this in C#?
Is there language agnostic solution?
Okay, Buffer.BlockCopy is significantly faster than copying data byte-by-byte, even after you eliminate bounds checking. Interestingly, it seems that Buffer.BlockCopy does in fact work on 2D arrays :)
So you might want to try something like this:
byte[,] source = new byte[1000, 100];
byte[,] dest = new byte[2048, 2048];
int offsetX = 100;
int offsetY = 20;
int width = source.GetLength(0);
int height = source.GetLength(1);
for (int y = 0; y < height; y++)
{
Buffer.BlockCopy
(
source, y * height,
dest, offsetX + dest.GetLength(1) * (y + offsetY),
width
);
}
Basically, assuming that the array is a byte array with X as the first index, and Y as the second, I go one row at a time, blitting the whole row from source to dest at once.
It seems much much faster than simply copying one byte at a time, so I assume that it does actually use DMA instead of using the CPU to copy the bytes.
Do note that this will only be faster if the rows are long enough. If you're copying a single column, it will probably be slower than just copying byte by byte. If you find yourself copying columns more often than rows (ie. width is usually less than height), you might want to think about inverting the coordinates, ie. swapping X and Y.
How can I do a color replace like the code below without using the if statement and instead use boolean algebra (or some other magic that will not introduce conditional logic)
The Problem (excuse the code):
private Image ReplaceRectangleColors(Bitmap b,
Rectangle rect,
Color oldColor,
Color newColor)
{
BitmapData bmData = b.LockBits(rect,
ImageLockMode.ReadWrite,
PixelFormat.Format24bppRgb);
int stride = bmData.Stride;
IntPtr Scan0 = bmData.Scan0;
byte red = 0;
byte blue = 0;
byte green = 0;
unsafe
{
byte * p = (byte *)(void *)Scan0;
int nOffset = stride - rect.Width *3;
for(int y=0; y < rect.Height; ++y)
{
for(int x=0; x < rect.Width; ++x )
{
red = p[0];
blue = p[1];
green = p[2];
if (red == oldColor.R
&& blue == oldColor.B
&& green == oldColor.G)
{
p[0] = newColor.R;
p[1] = newColor.B;
p[2] = newColor.G;
}
p += 3;
}
p += nOffset;
}
}
b.UnlockBits(bmData);
return (Image)b;
}
The problem I have is that if the image is huge this code gets executed many times and has poor formance. I know there has to be a way to substitute the color replacement with something much cleaner/faster. Any ideas?
Just to summarize and simplify, I want to turn
if (red == oldColor.R
&& blue == oldColor.B
&& green == oldColor.G)
{
red = newColor.R;
blue = newColor.B;
green = newColor.G;
}
into a bit operation that doesn't include an if statement.
There aren't any bitwise operations that will replace pixels of one colour with another for you. In fact, reading a pixel, applying a bitwise operation and writing back the results for every pixel will probably work out slower than reading a pixel and only doing any work on it and writing it back if it matches your target colour.
However, there are some things that can be done to speed up the code, with increasing levels of complexity:
1) The first thing you could do is not to read the 3 bytes before you do the compare. If you read each byte only as it is needed for the comparison, then in the case that the red byte doesn't match, there isn't any need to read or compare the Green/Blue bytes. (The optimiser may well work this out on your behalf though)
2) Use cache coherence by accessing the data in the address-order that it is stored in. (You're doing this by working on the scanlines by putting x in your inner loop).
3) Use multithreading. Break the image into (e.g.) 4 strips, and process them in parallel, and you should be able to get a "several times" speedup if you have a 4+ core processor.
4) You may be able to work several times faster by using a 32-bit or 64-bit value instead of four or eight 8-bit values. This is because fetching one byte from memory might take a similar time (give or take some cache coherence etc) to fetching an entire CPU register (4 or 8 bytes). Once you have the value in a register, you can do a single comparison (RGBA) rather than four (R, G, B, A bytes separately), and then a single write back - potentially as much as 4x faster. This is the easy case (for 32-bpp images), as they conveniently fit one-pixel-per-int, so you can use a 32-bit integer to read/compare/write an entire RGBA pixel in a single operation.
But for other image depths you will have a much harder case, as the number of bytes in each pixel will not exactly match the size of your 32-bit int. For example, for 24bpp images, you will need to read three 32-bit dwords (12 bytes) so that you can then process four pixels (3 bytes x 4 = 12) on each iteration of your loop. You will need to use bitwise operations to peel apart these 3 ints and compare them to your 'oldcolour' (see below). An added complication is that you must be careful not to run off the end of each scanline if you are processing it in 4-pixel jumps. A similar process applies to using 64-bit longs, or processing lower bpp images - but you will have to start doing more intricate bit-wise operations to pull the data out cleanly, and it can get pretty complicated.
So how do you compare the pixels?
The first pixel is easy.
int oldColour = 0x00112233; // e.g. R=33, G=22, B=11
int newColour = 0x00445566;
int chunk1 = scanline[i]; // Treating scanline as an array of int, read 3 ints (12 bytes)
int chunk2 = scanline[i+1]; // We cache them in ints as we will read/write several times
int chunk3 = scanline[i+2];
if (chunk1 & 0x00ffffff == oldColour) // read and check 3 bytes of pixel
chunk2 = (chunk2 & 0xff000000) | newColour; // Write back 3 bytes of pixel
The next pixel has one byte in the first int, and 2 bytes in the next int:
if ((chunk1 >> 24) == (oldColour & 0xff)) // Does B byte match?
{
if ((chunk2 & 0x0000ffff) == (oldColour >> 8))
{
chunk1 = (chunk1 & 0x00ffffff) | (newColour & 0xff); // Replace B byte in chunk1
chunk2 = (chunk2 & 0xffff0000) | (newColour >> 8); // Replace G, B bytes in chunk2
}
}
Then the third pixel has 2 bytes (RG) in chunk2 and 1 byte (B) in chunk3:
if ((chunk2 >> 16) == (oldColour & 0xffff))
{
if ((chunk3 & 0xff) == (oldColour >> 16))
{
chunk2 = (chunk2 & 0x0000ffff) | (newColour << 16); // Replace RG bytes in chunk2
chunk3 = (chunk3 & 0xffffff00) | (newColour >> 16); // Replace B byte in chunk3
}
}
And finally, the last 3 bytes in chunk3 are the last pixel
if ((chunk3 >> 8) == oldCOlour)
chunk3 = (chunk3 & 0x000000ff) | (newColour << 8);
... and then write back the chunks to the scanline buffer.
That's the gist of it (and my masking/combining above may have some bugs, as I wrote the example code quickly and may have mixed up some of the pixels!).
Of course, once it works, you can then optimise it a load more - for example, whenever I compare stuff to parts of the oldColour (e.g. oldColour >> 16), I can precaclulate that constant outside the entire processing loop, and just use an "oldColourShiftedRight16" variable to avoid recalculating it on every pass through the loop. THe same goes for all the bits of newColour that are used. Potentially you may be able to make some gains by avoiding writing back the values that haven't been touched, too, as many of your pixels probably won't match the one you want to change.
So that should give you some idea of what you were asking for. It's not particularly simple, but it's a great deal of fun :-)
When you've got it all written and super-optimised, then the final step is to throw it away and just use your graphics card to do the whole thing a bazillion times faster in hardware - but let's face it, where's the fun in that? :-)
I wrote a project recently where I did color manipulation on a pixel per pixel basis. It had to run fast as it would update while you moved a mouse cursor around.
I started with unsafe code but I don't like unsafe code and so changed to safe territory and when I did, I had the speed issues you had but the resolution wasn't changing conditional logic. It was designing better algorithms for the pixel manipulation.
I'll give you an overview of what I did and I'm hoping it can get you where you want to be because it's really close.
First: I had multiple possible input pixel formats. Due to that I couldn't assume the RGB bytes were at specific offsets or even a static width. As such, I read the info from the passed in image and return a "color" that represents the sizes of each field:
private System.Drawing.Color GetOffsets(System.Drawing.Imaging.PixelFormat PixelFormat)
{
//Alpha contains bytes per color,
// R contains R offset in bytes
// G contains G offset in bytes
// B contains B offset in bytes
switch(PixelFormat)
{
case System.Drawing.Imaging.PixelFormat.Format24bppRgb:
return System.Drawing.Color.FromArgb(3, 0, 1, 2);
case System.Drawing.Imaging.PixelFormat.Format32bppArgb:
case System.Drawing.Imaging.PixelFormat.Format32bppPArgb:
return System.Drawing.Color.FromArgb(4, 1, 2, 3);
case System.Drawing.Imaging.PixelFormat.Format32bppRgb:
return System.Drawing.Color.FromArgb(4, 0, 1, 2);
case System.Drawing.Imaging.PixelFormat.Format8bppIndexed:
return System.Drawing.Color.White;
default:
return System.Drawing.Color.White;
}
}
For example purposes, let's say that a 24-bit RGB image is the source. I didn't want to change alpha values as I'm going to blend a color in to it.
Thus, R is at offset 0, B is at offset 1 and G at offset 2 and each pixel is three bits wide. This I create a temporary Color with this data.
Next, since this is in a custom control, I didn't want flickering so I overrode the OnPaintBackground and turned it off:
protected override void OnPaintBackground(System.Windows.Forms.PaintEventArgs pevent)
{
//base.OnPaintBackground(pevent);
}
Finally, and here's the part that gets to the crux of what you're doing, I draw a new image on each OnPaint (which is triggered as a mouse moves because I "Invalidate" it in the mouse move event handler)
Full code - before I call certain sections out ...
protected override void OnPaint(System.Windows.Forms.PaintEventArgs pe)
{
base.OnPaint(pe);
pe.Graphics.FillRectangle(new System.Drawing.SolidBrush(this.BackColor), pe.ClipRectangle);
System.Drawing.Rectangle DestinationRect = GetDestinationRectangle(pe.ClipRectangle);
if(DestinationRect != System.Drawing.Rectangle.Empty)
{
System.Drawing.Image BlendedImage = (System.Drawing.Image) this.Image.Clone();
if(HighlightRegion != System.Drawing.Rectangle.Empty && this.Image != null)
{
System.Drawing.Rectangle OffsetHighlightRegion =
new System.Drawing.Rectangle(
new System.Drawing.Point(
Math.Min(Math.Max(HighlightRegion.X + OffsetX, 0), BlendedImage.Width - HighlightRegion.Width -1),
Math.Min(Math.Max(HighlightRegion.Y + OffsetY, 0), BlendedImage.Height - HighlightRegion.Height -1)
)
, HighlightRegion.Size
);
System.Drawing.Bitmap BlendedBitmap = (System.Drawing.Bitmap) BlendedImage;
System.Drawing.Color OffsetRGB = GetOffsets(BlendedImage.PixelFormat);
byte BlendR = SelectionColor.R;
byte BlendG = SelectionColor.G;
byte BlendB = SelectionColor.B;
byte BlendBorderR = SelectionBorderColor.R;
byte BlendBorderG = SelectionBorderColor.G;
byte BlendBorderB = SelectionBorderColor.B;
if(OffsetRGB != System.Drawing.Color.White) //White means not supported
{
int BitWidth = OffsetRGB.G - OffsetRGB.R;
System.Drawing.Imaging.BitmapData BlendedData = BlendedBitmap.LockBits(new System.Drawing.Rectangle(0, 0, BlendedBitmap.Width, BlendedBitmap.Height), System.Drawing.Imaging.ImageLockMode.ReadWrite, BlendedBitmap.PixelFormat);
int StrideWidth = BlendedData.Stride;
int BytesPerColor = OffsetRGB.A;
int ROffset = BytesPerColor - (OffsetRGB.R + 1);
int GOffset = BytesPerColor - (OffsetRGB.G + 1);
int BOffset = BytesPerColor - (OffsetRGB.B + 1);
byte[] BlendedBytes = new byte[Math.Abs(StrideWidth) * BlendedData.Height];
System.Runtime.InteropServices.Marshal.Copy(BlendedData.Scan0, BlendedBytes, 0, BlendedBytes.Length);
//Create Highlighted Region
for(int Row = OffsetHighlightRegion.Top ; Row <= OffsetHighlightRegion.Bottom ; Row++)
{
for(int Column = OffsetHighlightRegion.Left ; Column <= OffsetHighlightRegion.Right ; Column++)
{
int Offset = Row * StrideWidth + Column * BytesPerColor;
if(Row == OffsetHighlightRegion.Top || Row == OffsetHighlightRegion.Bottom || Column == OffsetHighlightRegion.Left || Column == OffsetHighlightRegion.Right)
{
BlendedBytes[Offset + ROffset] = BlendBorderR;
BlendedBytes[Offset + GOffset] = BlendBorderG;
BlendedBytes[Offset + BOffset] = BlendBorderB;
}
else
{
BlendedBytes[Offset + ROffset] = (byte) ((BlendedBytes[Offset + ROffset] + BlendR) >> 1);
BlendedBytes[Offset + GOffset] = (byte) ((BlendedBytes[Offset + GOffset] + BlendG) >> 1);
BlendedBytes[Offset + BOffset] = (byte) ((BlendedBytes[Offset + BOffset] + BlendB) >> 1);
}
}
}
System.Runtime.InteropServices.Marshal.Copy(BlendedBytes, 0, BlendedData.Scan0, BlendedBytes.Length);
BlendedBitmap.UnlockBits(BlendedData);
//base.Image = (System.Drawing.Image) BlendedBitmap;
}
}
pe.Graphics.DrawImage(BlendedImage, 0, 0, DestinationRect, System.Drawing.GraphicsUnit.Pixel);
}
}
Going through the code here are some explanations...
System.Drawing.Image BlendedImage = (System.Drawing.Image) this.Image.Clone();
It is important to draw to an offscreen image - this creates one such image. Otherwise, the drawing will be much slower.
if(HighlightRegion != System.Drawing.Rectangle.Empty && this.Image != null)
HighlightRegion is a RECT that holds the area to "mark off" on the source image. I have used this to mark off image regions of 4 Million pixels and it still runs fast enough to be "real time"
Some code below is used because a user might be scrolled over or down on the image so I modify my destination by their scrolling amount.
Below that, I cast the IMAGE to a BITMAP and get the before-mentioned Color info which I'll need to start using now. Depending on what you're doing you might want to cache that instead of getting it each time.
System.Drawing.Bitmap BlendedBitmap = (System.Drawing.Bitmap) BlendedImage;
On my control, I exposed two Color properties - SelectionColor and SelectionBorderColor - so that my regions still have a nice border with them. Part of my speed optimization was to pre-cast these to bytes as I'll be doing bitwise operations in a moment.
You'll see a comment in the code "White not supported" - in this case, the "White" is the "Fake Color" we use to store our bit widths. I used "White" to mean "I can't operate on this data"
The next line establishes that indeed each color is one bit because they might not be depending on our target color format by subtracting the R and G offset. Note that if you cannot garauntee that your G follows your R then you'll need to use something else. In my case, it was garaunteed.
Now where the part you're really looking for starts. I use a LockBits to get the bit data. After that, I use the data to finish setting up some pre-loop variables.
And then, I copy the data to a byte array. I'm going to loop through this byte array, change the values and then copy it's data back to the BITMAP. I was working on the BITMAP directly before thinking that since it's offscreen it would be just as fast as working with a native array.
I was wrong. Performance profiling proved it to me. It's faster to copy everything to a byte array and work within that.
Now the loop starts. It goes row by row, column by column. Offset is a number telling us where in the byte array we are in terms of "current pixel".
Then, I blend 50% or I draw a border. Note that for each pixel I have not only an IF statement, but also OR checks.
And it's still fast as blazes.
Finally, I copy back and unlock the bits. And then copy the image to the onscreen surface.