Research On CMOS Image Sensor Noise Suppression

Digital cameras usually employ a single CMOS sensor whose surface is covered with a color filter array(CFA).The CFA limits each sensor pixel just sampling one of the three primary color values (red, green or blue). The acquired image is then processed through an imaging pipeline of interpolating, denoising and color correction. Sensor noise deteriorates the digital camera outputs. The algorithms used in each stage of the image pipeline are complex, several of them nonlinear and the effect on the noise is often too complicated to describe mathematically at the output.First, the problems of total least squares method for interference with multiplicative noise of image sensors are studied in this thesis. Second, the thesis investigates the effects of the order of interpolating and denoising process on images and image noise. The study focuses on commonly adopted four denoising algorithms: total least squares, block matching and 3D filtering, bilateral filtering, linear filtering and three interpolating algorithms, namely bilinear interpolation, adaptive homogeneity, projection onto convex sets. However, interpolating process suffer from bad input noise while correlated noise passed from interpolating operation degrades the output of denoising algorithms. Thus, it is reasonable to consider doing denoising before applying interpolating method. In addition, a joint interpolating and denoising algorithm which combines these two procedures systematically into a single operation is studied. Finally, research on color balance is given.Through establishing of CMOS image sensor noise analysis model, selecting the RAW image as input, RGB image as output, CMOS image sensor noise suppression is studied on the modular method. To test and evaluate the performance for each imaging program the modules are developed to assess the image quality and noise characteristics, these modules include interpolatiing module, denoising module, joint interpolating and denoising module, image noise assessment module, data analysis module.The creative works of this thesis include:for the first time, the total least squares method based on image noise reduction is optimized, focusing on the relationships between image block selection, window selection and estimation performance, and successfully applied to denoising on CMOS image sensors after interpolation.The effects on the imaging performance from the different orders of interpolating and denoising process are studied, including three treatment orders: denoising after interpolating, interpolating after denoising, joint interpolating and denoising at the same time.Three effective methods for the RAW image with noise are proposed and studied.The typical combination of color balance algorithm is studied, and the gray world method, mean and standard deviation method is optimized.