Studying And Implementation Of Image Signal Preprocessing Based On FPGA

Image Signal Preprocessing(ISP) is the algorithm of image sensors, which is used to produce digital images of standard format. ISP plays an important part of image sensors, affecting the image quality, and has great economic and political significance. ISP is a set of many algorithms. This paper selects the most four basic algorithms to be studied, building a simple ISP system. The selected algorithms are Demosaic, Auto White Balance, Gamma Correction, and image coding and decoding. All four algorithms are optimized in this paper considering their drawbacks and FPGA’s features.Demosaic algorithm fills Bayer format images to RGB images, complementing the 3 channels of color. The Demosaic algorithm selected in this paper is traditional gradient interpolation method, whose transition is not smooth enough. To solve this problem, this paper optimizes the algorithm with the features of FPGA, expanding the scope of sample in gradient calculation and Green Channel interpolation.Auto White Balance(AWB) could correct the influence of light source, adjusting the visual impact to standard light source. The dynamic threshold is selected to achieve AWB. During the implementation, a pixel of the maximum brightness is needed, which is sensitive to noise. At this point, we introduce a filter, substituting the maximum brightness with the weighted sum of the maximum 8 brightness.Gamma Correction is used to increase the contrast of the part of mid luminance, making images suitable for human eyes. Traditionally, FPGA implementations use straight lines to approach Gamma curve. This method makes a big error in the big curvature part. In this paper, we replace the straight lines with Taylor series twice, and the accuracy is highly improved, achieving a much better result.There are various to code and decode images. In consideration of FPGA, we select DPCM Coding. Traditional DPCM Coding could lead to error diffusion, and has a small scale of difference coding. To fix these 2 points, we come up this a totally new ADPCM Coding, which not only stop the error diffusion, but also increase the scale of difference coding, decreasing the losing of compression.All four algorithms are simulated and compared, and finally implemented with FPGA. A simple ISP system is built and simulated, and is proved advanced and pratical.