Compressive image sensor architecture with on-chip measurement matrix generation

Abstract

A CMOS image sensor architecture that uses a cellular automaton for the pseudo-random compressive sampling matrix generation is presented. The image sensor employs in-pixel pulse-frequency modulation and column wise pulse counters to produce compressed samples. A common problem of compressive sampling applied to image sensors is that the size of a full-frame compressive strategy is too large to be stored in an on-chip memory. Since this matrix has to be transmitted to or from the reconstruction system its size would also prevent practical applications. A full-frame compressive strategy generated using a 1-D cellular automaton showing a class III behavior neither needs a storage memory nor needs to be continuously transmitted. In-pixel pulse frequency modulation and up-down counters allow the generation of differential compressed samples directly in the digital domain where it is easier to improve the required dynamic range. These solutions combined together improve the accuracy of the compressed samples thus improving the performance of any generic reconstruction algorithm.