Real Time Digital Image Processing Using Point Operations in Multithreaded Systems

This paper deals with real-time digital image processing using multi-threading for efficient and reliable real-time output. Two types of classes are dealt with: class A where processing is based on single pixel operations and class B where output is based on neighborhood pixel values, which is dedicated to convolution operations. Partitioning an image into sub-images to process it using multiple threads is not as simple as it looks. One issue is of information loss, which occurs during partition, especially at boundary pixels. Calculation of optimal size of sub-images is another issue. Whereas, writing code for gaining maximum performance via multi-threading is also cumbersome. In this paper, issue of how loss affects the speed of output with limited or desired number of resources is tackled. The aim is to reduce the information loss and improve performance. A generic multi-threaded API, to handle different image operations, is tested on images of different sizes and performance is compared with single-threaded approach. Experimental results demonstrate better execution time without data loss.