A robust algorithm for morphological, spatial image-filtering and character feature extraction and mapping employed for vehicle number plate recognition

This paper illustrates the implementation of a character feature extraction, classifier and an effective mapping algorithm applied on images of vehicle license plates, using nonlinear feature extraction and spatial mapping. A novel algorithm for low-level visual feature extraction has been presented, highlighting non-linear morphological operations with structuring, filtering, erosion and dilation techniques. Filtering and enhancement of raw image for key feature extraction and recognition by edges by sobel, canny and Fuzzy logic, convolution, median filtering and scaling image for boosting visual spatial features has been performed. Template matching using thresholding by effective feature mapping and semantic feature matching methods with pre-fed datasets has been accomplished as an application to image computing. The results obtained are tabulated and an accuracy of 79.30% using the above mentioned algorithm is reported. The images henceforth have been exposed to standard channel noises and thereafter compared for loss of information and overall structure. Comparison methodologies used are MSE and PSNR values and Structured Similarity Index (SSIM). The accuracy of results thus obtained are reported, suggesting the robustness and effectiveness of the algorithm. Image acquisition is accomplished by a remotely controlled android platform based device integrated with the MATLAB platform in real-time. The acquired number-plate data is transferred into a text file for creating a vehicle index log, critical for security systems.