PROJECT TITLE :

DERF: Distinctive Efficient Robust Features From the Biological Modeling of the P Ganglion Cells - 2015

ABSTRACT:

Studies in neuroscience and biological vision have shown that the human retina has sturdy computational power, and its information illustration supports vision tasks on both ventral and dorsal pathways. During this paper, a brand new native image descriptor, termed distinctive economical strong options (DERF), is derived by modeling the response and distribution properties of the parvocellular-projecting ganglion cells in the primate retina. DERF features exponential scale distribution, exponential grid structure, and circularly symmetric perform distinction of Gaussian (DoG) used as a convolution kernel, all of which are according to the characteristics of the ganglion cell array found in neurophysiology, anatomy, and biophysics. Yet, a replacement clarification for local descriptor design is presented from the perspective of wavelet tight frames. DoG is naturally a wavelet, and therefore 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 once we modulate the parameters of our descriptor to create the frame tighter, the performance of the DERF descriptor improves accordingly. This is verified by coming up with a tight frame DoG, that results in a lot of better performance. In depth experiments conducted in the image matching task on the multiview stereo correspondence information set demonstrate that DERF outperforms state of the art methods for both hand-crafted and learned descriptors, whereas remaining robust and being a lot of faster to compute.