The Technology And Research On Location Of Character Image And Feature Point's Matching

Character is the important content information in images and video frames. It's very important for image comprehending, retrieving and browsing that automatic locating and extracting the text regions in image and video frame, then extracting the character region from images, and last recognizing characters by technique of feature point's matching.The identification process of character in character images can be divided into two parts: character region's location and single character's recognition. In this paper, the technology of recognition on character regions is studied and proposed. The major work implemented in this paper is presented as follows:1) Character region's location. This paper uses wavelet transformation in character image and morphology to segment video image character. The experimental result shows that this method can extract character regions effectively, such as license plates and text. Similarly, it also has good effect on slope character regions.2) Character's segmentation and recognition. This paper uses the labeling algorithm and the method of connected area to segment characters, and then realizes character's recognition during the method of feature point's extraction and matching in character. The experimental result shows that the above method is satisfied when we use it to solve the problem of character's segmentation and recognition.3) In the stage of character region's location, this paper proposes a new template detection algorithm based on improved Minimum Far Distance. This algorithm reduces the tested calculation by the method that replaces dynamical adjusted pixel similarity threshold with the threshold which has no change in the traditional method. The experimental result shows that this algorithm is no more than the traditional algorithm in the aspects of the numbers of pixel's computation and run-time.4) In the stage of character recognition, this paper proposes a new feature point's matching algorithm based on improved Hopfield neural network. This approach improves Hopfield network by the addition of the method that input-output function of network, the difference measurement between two points and selecting target points. The experimental result shows that compared with the traditional Hopfield network, our improved algorithm have a higher matching accuracy, a better matching effect on image feature point, and runs faster than traditional network.