Mobile type of automatic identification system for a car plate

An automatic identification method for a cap-plate comprising a photographing apparatus adapted to take a picture of the car-plate while under still or moving condition, an image-processing CPU, and means for transmitting said picture into said image-processing CPU for converting said picture into a digital image comprising a plurality of pixels arranged in a plurality of columns by a plurality of rows, each pixel being represented by a gray level, and for automatic identification of characters contained in said car-plate;

• wherein said automatic identification of characters performed by said image-processing CPU comprises the steps of:
• (a) finding the position of the car-plate in said image and extracting a partial image corresponding to said car-plate using an appropriate methodology;
• (b) calculating a vertically projected array of said partial image containing said car-plate by summing the gray levels of all the pixels in a same column, and storing the result so calculated into a one-dimensional array X;
• (c) calculating a first differential vertically projected array dx by performing a first difference process of said partial image containing said car plate as obtained in step (a) and summing values of calculated differences in each column to form said array dx, said first difference process is formed by subtracting from the gray level of each pixel thereof by the gray level of a neighboring pixel in the next column but on the same row and taking the absolute value of the difference;
• (d) establishing a fuzzy rule base by forming a membership function which is a function of X and dx obtained in steps (b) and (c) above;
• (e) using a max-min inference procedure in accordance with said fuzzy rule base to select a plurality of best dividing points to thereby frame and extract every individual character contained in said car-plate;
• (f) converting each extracted character into a plurality of pixels, which are then grouped into three pixel groups:
• a full shape group, a top shape group and a bottom shape group, wherein said full shape group contains all the pixels corresponding to said extracted character, and said top shape and bottom shape groups each containing only a portion of said pixels;
• (g) forming a four-layer neural network containing four layers:
• an input layer, a character structure layer, an implicit layer, and an output layer, said input layer containing portions corresponding to said full shape group, said top shape group and said bottom shape group obtained in step (f);
• and
• (h) using said neural network from step (g) to identify all said characters extracted in step (e).