A Research And Application On Image Detection And Recognition Technology

Image processing, analysis and machine vision represent an exciting and dynamic part of cognitive and computer science. This sequence of operations-image capture, early processing, region extraction, region labeling, high-level identification, qualitative /quantitative conclusion is characteristic of image understanding and computer vision problems.The object in the world can be divided into two kinds: natural object and man-made object. The man-made object can be defined by some simple geometrical units, but the natural object usually has complicated outline and structure. The methods to detect and recognize two kinds of object are different according to their difference feature. We research object detection and recognition technology by selecting building and human face as the samples which are typical in the normal life.The main contributions of this paper are as follows.Firstly, in the research of building detection and recognition, the methods of extracting lines in images are introduced in detail. We use lines to describe and explain buildings'outline. We use multi-features as evidences to decide whether the lines can link or not to extract lines, which reduces the possibility of wrong decision. Then a method of extracting outline and windows of building is presented by analyzing the lines.Secondly, a survey on the state of face detection on research based on systematic analysis of related papers is presents in this paper. We introduce this method of the AdaBoost based classifier presented by Viola in detail and detect faces using this method. A new face normalization method based on eyes location is presented. Firstly, we accurately locate the pupils of eyes in the face image according to the proportion relationship of face features and gray information. Then we normalize the rotation, scale and grayscale of face image. We recognize human face using the method based on embedded hidden Markov model (EHMM) that used the 2D-DCT coefficients as the observation feature. The experimental results show that this algorithm can improve the face recognition obviously.