Foreground object features extraction with GLCM texture descriptor in FPGA

Mateusz Komorkiewicz, Marek Gorgon

Abstract: Texture descriptors are a powerful tool for 2D scene features extraction. They can be computed for whole image or for regions of interests obtained from various object detection methods. In case when foreground objects mask obtained from background subtraction or optical-flow thresholding is to be used, the connected components analysis is needed first. The texture descriptor is then computed for each labeled segment in the second stage. This pipelined approach requires access to external memory to save the intermediate results and introduces high latency. In the article an FPGA based system is presented which is using a close integration of single pass connected components analysis module and GLCM (gray level co-occurrence matrix) texture descriptor computation block to obtain low latency, high speed foreground object feature extraction subsystem which is not requiring an external memory access.