SOFTWARE AND HARDWARE IMPLEMENTATION OF THE NEUROMORPHIC LGN BASED IMAGE PROCESSING AND FEATURE EXTRACTION

Abstract

The processing of the graphical data is popular methodology of obtaining important information. However, there is a major drawback: it typically requires large computational resources. The human brain is an excellent example of the efficient image processing hardware, due to the fact the biological visual system can allow to easily and quickly obtain the information of the world around, such as object identification and movement detection. In particular, there is one element of the biological visual system that has unique functionality in image processing, which is lateral geniculate nucleus (LGN). Ganglion cells, which are terminated in LGN, have high sensitivity to the image spatial intensity difference. This cell feature is used for pre-processing of the visual data before being modulated and relayed to the main processing module, visual cortex. As a result, the processing load on cortex is reduced, due to the pre-processing of the data. The aim of the project is to develop the algorithm for visual features extraction, such as edge detection, based on the structure and properties similar to the LGN, with a possibility of the hardware model implementation. Preliminary results show the edge detection property of the proposed method. Moreover, the measured performance is comparable to other popular edge detection techniques, even exceeding expectations to small extent in the noisy environment.