Parallel image segmentation using the L4 network

Each processing element in the outer plexiform layer of the fly's (Musca domestica) compound eye has a single copy of a monopolar cell called L4 whose function is still unknown. This paper proposes that L4 acts as an image segmentor receiving data from the L1 and L2 cell layer above it. The photoreceptor terminals R1 through R6 and L1 and L2 form a cartridge with shunting current inhibition that enhances contrast at the first synaptic contact. The photoreceptors feeding their respective terminals share a common optical axis. What was realized for this entire structure that led up to and included L1 and L2, was that it forms a single processing element which outputs a weighted function based on current shunting between the photoreceptor axon terminals around L1 and L2. Thus the basic biological 'algorithm' that this paper proposes for L1 and L2 involves a summation of the differences between a central reference cell (referred to as Rref) and the six neighboring photoreceptor terminals R1-R6 within the cartridge of L1 and L2. Simulation implementing this simple algorithm on a hexagonal packed matrix in Matlab, suggest that objects with even small differences in intensity with the background, can readily be distinguished from the background. This high pass filter biological model allows edge detection and segmentation, independent of scale in a parallel, modular fashion. Implementation of such a biological algorithm using analog circuitry forms a preprocessing unit that introduces virtually no delay in the processing of image information, in comparison to present DSP techniques, that require iterative approaches and costly computing time.