PARALLEL ALGORITHM AND OPTIMIZATION METHOD FOR CAMERA ISP

摘要

ISP is composed of a chain of various image processing algorithms such as denoising, white balancing, image enhancement, gamma correction, and the like. Most recent ISP algorithms having features of iteration and adaptivity output excellent image quality, but require a vast amount of computations and a lot of branches resulting from conditions. The ISP algorithms lack parallelism due to the selective operation processes, so the ISP algorithms are used in a VLIW processor and the like because the ISP algorithms are difficult to be used in an SIMD processor having a relatively efficient parallel structure. Thus, an optimization method for the ISP algorithms suitable for a processor having the parallel structure is needed. An optimization suitable for parallel processing may be carried out using the optimization method including making coefficients used in the algorithms integers, modifying a 2D filter to be separable, modifying the order of a used formula, simplifying image feature measures used for adjusting the adaptivity of each algorithm, approximating the algorithms, rearranging the algorithms, and the like. This facilitates improvement of image quality, control of the amount of computations, and real-time operations in various parallel platforms.;COPYRIGHT KIPO 2017