The Image Restoration Of Special Weather

Image acquisition can be easily affected by severe weather. Under heavy weather, the contrast ratio declines and the color of the images fades. These degraded images exert negative influence on video surveillance, target extraction, target tracking and other relevant fields. On this very note, this thesis mainly focuses on the degraded images which are resulted from severe weather, studies the reason for image degradation and on the basis of the existed restoration algorithms, provides several modified ones for degraded images affected by bad weather.The main tasks include:1. The restoration of fog-degraded images based on the modified dark channel prior. By applying the dark channel prior, the fog-degraded image becomes darker after the restoration process. In order to overcome this shortage and achieve restoration of the degraded images, the light intensity needs to be appropriately compensated and the transmissivity estimated by combining the dark channel prior and soft matting method.2. The restoration algorithm of multiple fog-degraded images based on the modified principle of polarization. The restoration process is achieved through studying the physical model of atmospheric scattering and the principle of polarization, adopting the least square method (LSM) to calculate the restoration parameters of fog-degraded images and optimizing these figures on a more reasonable and related basis.3. The restoration algorithm of multiple fog-degraded images based on the difference of the depth of field. This method mainly uses the relation between the depth of field and transmissivity, employs the LSM to get transmissivity in terms of multiple images, utilizes the figure of depth to fit into a curve and provides reasonable optimization processing for these parameters to ultimately achieve the restoration of degraded images.4. The restoration of snow-degraded images based on the physical model of snowflake imaging. The first step is to remove the snow by using dark channel prior after studying the physical model of snowflake imaging and the similarity between the snowflake imaging model and the atmospheric scattering model, then to remove the remaining snow by applying the median filter which is adaptive to the size of the snowflake.After studying of existing evaluation criterion of images and the main evaluation criterion of fog-degraded images, standard deviation and average gradient are selected as the criterion to evaluate the restoration performance and the effectiveness of the algorithms. Based on experimental observation, the algorithms listed in this thesis are capable of producing clearer images and thus deliver better results in terms of the restoration of degraded images.