| In recent years,with the development of computer information and the emergence of artificial intelligence, pattern recognition has become a new discipline rapidly. By making use of the principles of human vision, image processing technology uses digital image as the main processing object,and then extracts the important information with some algorithms serving for the pattern recognition. As the initial processing in intelligent recognization system, the detection accuracy and efficiency of head detection may have a direct impact on the final recognition results, thus, it has aroused people's great research interests and passions. Sharp-based head detection method analyses the features which can distinguish head from other objects, and then uses those features to confirm head areas. This method has been widely used because of its high accuracy and strong anti-interference, especially when the scene of intelligent application is volatile.The main work of this thesis is summarized as follows.Firstly, a method based on interframe difference and background difference was used to extract the motion objects, and Gaussian mixture model is utilized to update the background. This method can extract the moving object steadily, and provides the necessary guarantees for the following test.Secondly, on the basis of a thorough research in the current head detection algorithms, a head detection method based on curvature scale space is proposed. The algorithm assumes that there are two obvious concave points between head and neck contour, like shape "Ω", and uses the corner detection method to find the feature points on the head contour. The head of different angles are sampled so as to confirm the range of the rate of head sharp. Experiments show that the algorithm can detect the head of the image with low false detection rate. The method is compared with several existing algorithms under different situations. The results show that the method is reliable and has strong suitability. |
PDF Full Text Request