Research On The Image Recognition Based On Neural Network

The image recognition is developing in the last 20 years. The image recognition is become the main contest of the classification and description in some objects or process (regarding as the Image). The image recognition have many extensive research, such as the image recognition can recognize the cancer cell from the medical image; it can recognize the form, forest and lake from the remote sensing image; it can recognize the car which made the traffic accident from the traffic image and so on. This paper analyzes the traditional methods of the image recognition. Then it finds that the traditional methods have some localization which is bar for the development of the image recognition. At last a new kind of the image recognition is put forward to by this paper which make some improvement according to the neural network based on neural network.The research way for this paper according with the basic step of the image recognition is that the first, realizes the color image turn to the gray image and the binary image using the current formula; the second, removes the noise from the image using the smoothing filtering and median filtering and the sharp filtering; the third, gets a new edge detection method for the image segment which is the result that research the roberts operator and the sobel operator and the orthogonal operator, this method can make the image segment valid; the forth, analyses the influence of scale factor on moment invariant features under the state of discrete and adopts a kind of modified algorithm of moment invariants to abstract invariants of object; the fifth makes the certain of the hidden layer and the hidden unit of BP network and improves the network by modifying initial weight vectors which can make the network efficiently. The point of this paper is the image segment and the moment invariant feature and the neural network. Lastly this paper makes a program to realize the function.1. a new the edge detection methodThe image segment is the moment step for the image recognition. This paper analyzes the roberts operator and the sobel operator and the orthogonal operator, finds that the three operator make the certain initial weight vectors but this initial weight must not make certain in the actual circumstance. So this paper proposes a new kind of edge detection method which make use of the 3 X 3 template. This operator can judge the direction of the edge, and make certain the initial weight vectors according to the edge. The analysis of the configuration mechanism of the new operator and large numbers of experiments show that the new approach, which is a good supplement to edge detection, is feasible.2. a new the moment invariants on the extracting characteristicThe extracting characteristic is key of the image recognition. This paper analyzes the moment...