Research On Video Processing Key Techniques Of Improving Image Quality In Special Scenes

Images of outdoor scenes are usually degraded in special conditions, especially in haze, backlight and night conditions. The degraded images lose the contrast and color fidelity, it will affect the outdoor imaging system. Therefore, it is important to effectively improve the quality of images in special conditions. This thesis detailed analyze the reason of image degradation, research on the basic theories and key technologies, and make some improvement on the existing methods. The concrete research contents are as follows.1. After we detailed analyze the haze imaging model and the single image haze removal using dark channel prior, an improved and fast dehazing method is proposed. This method make some improvement in the algorithm complexity and application scene. The experimental results show that the presented method significantly improve the quality of hazy images, not only reduce the algorithm complexity, but expand the application of dark channel prior as well.2. For the real-time requirement of engineering applications, a real-time dehazing method is proposed. This method analyzes the reason of the high complexity of single image haze removal using dark channel prior, improve the transmission estimating method, greatly reduce the algorithm complexity. The experimental results show that this method not only effective, but also meet the real-time requirement of engineering applications.3. For wide dynamic range images, a method based on image fusion is proposed. This method uses original image to generate image sequences to avoid the difficulty of obtain image sequences of the same scene, and then using pyramid model to image fusion. The results of experiment show that the algorithm could obtain approximate effect of multiple images fusion.4. According to the conclusion that an inverted wide dynamic range image is similar to a haze image, we present a fast image enhancement algorithm for wide dynamic range images. The algorithm works by inverting the input image first and then applying an optimized image haze-removal algorithm based on physical model on the inverted image. Experiments show that the presented method enhances the visual effect of the wide dynamic range image effectively and fast, making it suitable for real-time applications.5. This thesis analyzes the experimental results from the efficiency and application of the algorithm, and combined with contrast, information entropy and no reference image quality assessment. It compared dehazing effect, algorithm complexity and objective quality evaluation value with different methods.6. This thesis builds a software system for image and video processing in special scenes, implement hazy image and video restoration processing, wide dynamic range image and video enhancement processing, parameter adjustment and saving results.