PROJECT TITLE :

DERF: Distinctive Efficient Robust Features From the Biological Modeling of the P Ganglion Cells

ABSTRACT:

Studies in neuroscience and biological vision have shown that the human retina has robust computational power, and its information representation supports vision tasks on each ventral and dorsal pathways. During this paper, a brand new native image descriptor, termed distinctive efficient sturdy options (DERF), springs by modeling the response and distribution properties of the parvocellular-projecting ganglion cells in the primate retina. DERF options exponential scale distribution, exponential grid structure, and circularly symmetric function distinction of Gaussian (DoG) used as a convolution kernel, all of that are per the characteristics of the ganglion cell array found in neurophysiology, anatomy, and biophysics. Furthermore, a brand new rationalization for local descriptor design is presented from the angle of wavelet tight frames. DoG is of course a wavelet, and the structure of the grid points array in our descriptor is closely related to the spatial sampling of wavelets. The DoG wavelet itself forms a frame, and after we modulate the parameters of our descriptor to make the frame tighter, the performance of the DERF descriptor improves accordingly. This can be verified by planning a good frame DoG, which leads to much higher performance. In depth experiments conducted in the image matching task on the multiview stereo correspondence knowledge set demonstrate that DERF outperforms state of the art ways for each hand-crafted and learned descriptors, while remaining strong and being abundant faster to compute.