An Image Indexing Method Based On Adaptive Noise Filter

With more and more digital imaging equipment coming into people's daily life, image data, as a representative type of mass data in the Internet, is growing rapidly. Digital images are widely used both in production and life. Most applications of digital image are based on the image index. Since the image dataset possess the properties of of high noisy data, high dimensional sparse, therefore, building accurate index for image dataset is the key problem in image processing. In this paper, a new image index approach which is based on noise filter and Info-Kmeans clustering is proposed to solve this problem.The main contributions of this paper are as follows:(1) We survey the technologies related to image index, and discuss their deficien-cies in dealing with noise and high dimensional sparse clustering.(2) Focusing on the noise problem in image datasets, we proposed an image noise reduction technology based on frequent pattern mining technology. Experimental re-sults reveal that the noise in the image datasets can be effectively eliminated via such technology, and hence the quality of image indexing is significantly improved.(3) We propose ASAIL algorithm by introducing Shannon entropy into Info-Kmeans in order to improve the performance on high dimensional, sparse datasets. We first divide the original dataset into K clusters, and in each cluster we label the most frequently appeared image as the name of this cluster. Finally, the names of total clusters form the index of the image dataset.(4) We test our image indexing approach using two different types of image datasets. The results show that the noise reduction method proposed in this paper elim-inates noise image effectively, and comparisons between with and without noise re- duction, illustrate a significant improvement on the accuracy of index. Moreover, com-pared with traditional clustering algorithm, ASAIL shows higher performance when dealing with high dimensional, sparse datasets.