Image Reconstruction to Identify License Plate Numbers in Low Resolution Videos

Abstract

Surveillance videos are useful as a source for extracting information in many areas such as crime investigations and monitoring o ences in public roads. In many occasions, required information cannot be extracted from these videos due to the poor quality of the video or due to the high distance from camera to the interested object. A multiple image super resolution based technique to improve license plate regions in low quality videos have been proposed in this dissertation. Properly registering images is the main challenge in super resolution methods. An iterative method to register images correctly using phase correlation based image registration techniques have been proposed. The robustness of the method can be seen in experimental results. The evaluation of the method was done based on an automatic number plate recognition system. The impact to the recognition rates of automatic number plate recognition system by using the proposed method is measured. Compressed, low quality, real tra c videos were used as the dataset. According to the results, a signi cant improvement of recognition rates achieved with the proposed method.