Image Thresholding Computation of Between-Class Variance in a Partial Parallel Structure

Parallel processing of thresholding based on image between-class variance (BCV) is studied in this paper. In parallel processing, a frame of image is divided into M sub-images with the same size. Computation of the normalized probability and moments of image is distributed to each of the PCs. However, the rest of the computation in the BCV-based algorithm is non-parallel in essence. Hence, partial parallel processing structures are proposed in this study. Computer time required for parallel and non-parallel thresholding algorithms is compared on a working platform. Most of computer time is used for computing the image normalized probability. It is effective to distribute such computational efforts to multiple processors to promote the processing speed. Unfortunately, transmission of data among the processors takes dominant computer time. It induces a limitation on the processing speed of the distributed system. Conclusions drawn from this study show that 50% of the computer time the BCV-based algorithm takes can be reduced by using 4 processors in a proposed parallel structure