How to speedup image processing by parallel computing or multicore computing?

Question

Arunas 2012-8-30

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/47025-how-to-speedup-image-processing-by-parallel-computing-or-multicore-computing

编辑： Vishal_R 2014-6-16

I have designed image processing systems, which takes the image, preprocess it and do data mining with classification three. The result (class number) of captured image is further transmitted by COM1 port.

Currently I can run 1 frame analysis per 0.1s. Unfortunately my Matlab uses only 1 CPU core of 8 available. Therefore current application is TOO slow for online usage.

How I could use the matlab’s multicore ability from main matlab window? With matlabpool(4), I could speed up calculations up to 4 times, i.e.

1. getting image snapshot.

2. sending to first available LAB.

3. Without sequential waiting of the result TAKE next snapshot -> send to second available LAB/worker. and so on..

4. Collecting classification results asynchronously and sending them to COM port.

So … How to do it? Any ideas?

Regards, Arunas

2 个评论
显示无隐藏无

Dan Hickstein 2013-10-7

This is exactly what I would like to do. Not quite sure how to do it though...

Ashish Uthama 2013-10-7

Most compute intensive parts of MATLAB are multi-threaded (i.e should use all the 8 cores). It might be worthwhile to use the profiler to identify bottlenecks and follow up on them to speed it up.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Vishal_R 2014-6-16

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/47025-how-to-speedup-image-processing-by-parallel-computing-or-multicore-computing#answer_141193

编辑：Vishal_R 2014-6-16

在 MATLAB Online 中打开

You can process your image in parallel using blockproc function.

blockproc(x, {32 32},"Use Parallel", true, func);

where x is your image matrix, {32 32} is the block size (you can increase the block size for better results), func can be any customized function you want to apply on your block.

Use parallel will spawn default numbers of workers. Another alternative would be to use this function on GPU for faster results as compared to CPU.

-Vishal