Parallel Processing Architecture of Intuitive Digital Image Indices Based on Open Sources

Abstract

Digital image indices present specific image properties and have been proposed for statistical analysis. A considerable number of indices are calculated from the threshold, index function and sum of gray levels in image pixels. The computational cost of the indices is usually high owing to the repeated operations on megapixels in an image. Thus, this study discussed a parallel processing architecture to accelerate the computation of intuitive indices using open sources. The architecture was designed with various pixel depths, image sizes, region-of-interest, masking, and utilization for various indices. A base platform for image handling was constructed using the OpenCV library. The architecture was built using the open sources of a GPU and a multicore CPU. Thresholded content, a common digital focus index, was applied to verify the architecture. The processing time was measured to investigate the acceleration performance using various resolutions of industrial cameras. The architecture using the GPU and the multicore CPU decreased the computational cost and enabled real-time processing even for a large image.