Image Retrieval With Pulse Coupled Neural Classifier Based On Gabor Texture Features

Content based image retrieval (CBIR) is emerging as an important research area with great diversity of applications. Texture feature is very important information about CBIR, and Pulse coupled neural network (PCNN) is a new generation of artificial neural networks and is powerful in data processing. Some techniques of CBIR will be introduced in this paper. The two important issues, texture feature extraction and texture matching including matching strategy and matching method, are studied in great detail in this paper, where the texture feature of an image is extracted by applying a set of wavelet filters to the image, and the texture matching approach is divided into texture clustering and texture classification, the former as an unsupervised learning while the latter a supervised learning. An image index structure is created, for the real-time retrieval of the images in a large-scale texture image database. In order to get a high texture retrieval performance, PCNN is applied for clustering and classification because of its powerful adaptively in both clustering and classification. The performance of the method proposed is evaluated on the well-known Brodatz texture image database, showing that it is very effective and efficient in both retrieval rate and retrieval time, which are very suitable to be applied to large scale texture image database.