Research On Feature Extraction Algorithms In Chinese Costume Patterns

With the rapid development of economy, our country’s status in the world is higher and higher. More and more people begin to pay close attention to all aspects of China, and culture plays an important role. National culture is the priority, because it represents the development of the whole nation. An important carrier of national culture is national costume pattern. Only when contemporary scholars make a thorough study to national costume pattern can they inherit and carry forward our national culture better. The precondition of studying costume pattern is to classify patterns effectively, and the key of the classification is feature extraction. What kinds of feature extraction algorithms can be chosen and how to improve them to get better results will be discussed in this paper.To cure the above problems, this topic is supported by National Natural Science Foundation project(NO.61201118). The main content of the topic is as follows:(1) Research on application of feature extraction algorithms in Chinese costume patternsWe employ the image processing technology to extract pattern features of Chinese traditional costume in this paper. The representative algorithms of image feature extraction are SIFT and SURF which can deal with image rotation, scale changing and noise interference. We select one pattern of Qing dynasty court costume as the research object. The SIFT and the SURF are used to extract features of the image respectively, and then the BBF algorithm is used to determine the matching pairs of the feature points. At last the correct matching rate and the consuming time are calculated. The experimental results show that the SIFT and the SURF can maintain higher feature matching rate under conditions of image rotation, scale changing and noise interference.(2) Research on application of feature extraction algorithms based on RANSAC in Chinese costume patternsIn order to improve the correct matching rate of images, we combine the feature extraction algorithm and RANSAC in this paper. Firstly, the SIFT and the BRISK are used to extract match features of the images respectively, and then the RANSAC algorithm is used to remove part error matching points. At last the correct matching rate and the consuming time are calculated. The experimental results show that the two algorithms can maintain higher feature matching rate under conditions of image rotation, scale changing, shear transformation and noise interference, and the time complexity of the BRISK is lower.(3) The improved feature matching algorithm based on RANSACWe can see that the correct matching rate can be improved by combining SIFT and RANSAC from the above chapters. If we want to improve the correct matching rate further, we can make some improvements on the basis of the original algorithm. Considering the order of feature points on the characteristic dimension, we reduce error matching points by sorting and matching method to reduce the error matching rate. First, we apply RANSAC to SIFT and obtain the initial matching points. Secondly, we sort the matching points according to the abscissa and ordinate respectively to remove the inconsistent points. Third, we compute the centroid of the two images respectively and sort the distance between matching points and centroid to remove the inconsistent points. Last, we make the rest of the matching points as the final correct matching points. The experimental results show that under conditions of scale changing, noise interference, image rotation and shear transformation, the correct matching rate of the improved algorithm us enhanced in some degree.