DSP-Based Network Camera Image Pre-processing Technology

With the development of CPU and image sensor, video surveillance technology is booming, and IP network camera is more and more popular. The image quality is the core competitive power of cameras and also the fundamental of video analysis. High quality images need not only high-performance hardware, but efficient image pre-processing algorithms as well.The main task of this thesis is to study and optimize the camera image pre-processing algorithms. Realization and verification is done on the latest DM8127platform as well.Firstly, this thesis analyzed the camera image capture model. And combined with the factors of image quality, this thesis proposed the optimization target as guidance of image pre-processing algorithm study and optimization.Secondly, this thesis discussed camera3A algorithms. In the auto focus part, performance of each focus measurement function is evaluated, and an improvement search algorithm is proposed based on the specificity of auto focus, which uses various searching steps in different regions and is proven to be efficient in searching. In the auto exposure part, blur influence of aperture is analyzed and a new evaluation function is proposed. Auto exposure optimization is done by maximum the proposed function. The new auto exposure method is implemented on hardware and the experiment results show that this method is efficient in finding the best aperture when considering both blur and noise. In the auto white balance part, improvement is done base on gray world algorithm. The image is cut into regions to find the gray parts of image which is used in the calculation of light estimation. Iteration adjustment on camera parameters is performed based on the estimation results. Experiment results show that this algorithm has an efficient performance.At last, analysis and experiment of image processing modules and the integrated optimization is completed on DM8127platform. The best configuration is applied adaptively to achieve best performance. The system is implemented on DM8127software framework and proved to have a degree of improvement in subjective visual compared with original image.