AP Clustering Visual Dictionary PLSA Its Capacity To Obtain And Evaluate The Scene Classification

Scene classification is image understanding research focus, it is one of the important directions of human understanding of image semantic meaning. In recent years, scene classification in medical image recognition, computer vision, image retrieval and other fields have important applications. With the progress of time, the number of digital images of human output is growing. The method that relies on manual tagging approach for image acquisition of image annotation features and classification, can not satisfy human needs, therefore, automation Research scene classification is becoming increasingly important. Currently, the general scene classification processes are:image feature extraction, building visual dictionary, select classification algorithm. It is important to build a visual dictionary. We use AP clustering algorithm to automatically obtain a visual dictionary to build capacity, compared to the classical method to obtain a large number of tests, and efficiency is improved. In addition, the paper selects PLSA scene classification algorithm. The main work is as follows:First, the article describes the background knowledge scene classification and their meaning for a long time research, analyzes the clustering algorithm research status and gets a visual dictionary capacity.Second, this paper presents the basic flow chart scene classification. Get introduced SIFT (Scale-invariant feature transform, Scale-invariant feature transform) feature of the process, K-means clustering algorithm and AP clustering algorithm, and also analyzed the clustering results and compare randomly generated 100 data points their advantages and disadvantages. Meanwhile, PLSA algorithm describes in detail.Third, the paper studies several SIFT feature extraction method and experimental analysis of the advantages and disadvantages of using a uniform grid sampling and random sampling method to extract SIFT features images, and finally come to the uniform sampling method is more suitable for this article Research on scene classification.Fourth, this paper uses K-means clustering algorithm and AP poly tip algorithm to obtain a visual dictionary capacity, then SIFT scene classification algorithms, and analyze the advantages and disadvantages of the two methods. The classic use of K-means clustering algorithm requires a lot of testing before they can get a code book, AP clustering algorithm can automatically get a code book, greatly improving the efficiency of scene classification.Fifth, the paper analyzes the experimental results. Firstly, the use of K-means cl ustering algorithm to obtain code book and the use of two sets of experiments AP cl ustering algorithm automatically get code book and image of the scene classification. The results show that the AP clustering algorithm to get the visual scene classificati on capacity than K-means clustering algorithm to obtain the recognition rate and effic iency of the visual scene classification capacity to be higher. Then analyzed the resul ts of different topics k PLSA scene classification, and make a graph of their respecti ve experiments recognition rate, we can see from the figure, the number was 20 whe n the subject can get higher recognition rate.Sixth, the work done to make a summary of this article, and gives a follow-up study suggests.