Using super resolution to enhance license plates recognition accuracy

License Plate Recognition (LPR) has become one of the most widely used applications. One of the main factors affecting the accuracy of LPR is the quality of the image containing the plate. In case of video-based solutions, many frames are captured for the same car. Super Resolution (SR) techniques can be used for enhancing LPR accuracy by constructing one high resolution image from a number of low resolution images that belong to the same car. However because of the nature of the detected object (car plate), motion analysis as well as object perspective correction need to be taken into consideration when applying the SR algorithm over the provided series of video frames containing the moving car, in order to enhance the resolution of the plate area. This paper aims at providing an implementation for a SR algorithm where input frames are put together into one high-resolution image after detecting the license plate and removing the noisy frames to provide one clear and focused image as an output. OpenCV library was modified to support this implementation. This implementation enhanced the accuracy of an LPR solution by approximately 7% enhancement while trying to not increase the computational complexity of the original solution.