Research And Application Of Image Classification Method With Integrated Multi-feature By Adaboost Algorithm

With the rapid development of the network, image information has become a major part of the network information. However, the explosive growth of the images make people find their wanted pictures harder and harder. So the users enter the "Developed information, the lack of knowledge " cycle. This phenomenon promotes the generation of new techniques. The image classification technique is one of the methods. Image classification can automatically assign the appropriate category based on the image features. This method can greatly reduce the workload of human resource and improve the image retrieval’s accuracy. This thesis will introduce a method to class images by images’ multi-feature using Adaboost algorithm.The main contributions and innovations of this dissertation are as follows:This thesis presents a method using a variety of features for image classification. There are many features of an image. And the same type of images have a certain similarity in terms of the image features. Color feature, texture feature, shape feature, spatial relationships features are commonly used in image processing. One feature can only display one-sided of an image. Different image features have the different classification’s results. Therefore, in order to decribe an image much more accurate, the image features can be integrated to decribe the images. In this thesis, we use the color histogram method to extract color feature, the Wavelet Transform and GLCM method to extract texture feature and the Wavelet denosing and Hu moment invariants method to extract shape feature. The three features are integrated into one vector for classification.In this thesis, Adaboost algorithm is used as the image classification algorithm. Knn algorithm is used as the weaker classifier training algorithm. Adaboost is a iterative algorithm. By the increase of the iterations, it can effectively improves the accuracy of image classification.In this thesis, image classification method is integrated color, shape, texture features by Adaboost algorithm. Compare the experiment result with the use of a single image features and the use of knn algorithm to classify the images. The experimental result shows that the image classification integrated multi-feature by Adaboost algorithm has a better classification performance.