关键词: 基于内容的图像检索（CBIR）, 脉冲耦合神经网络（PCNN）, 交叉皮层模型（ICM）, 特征提取

Abstract: To increase the validity of texture feature extraction in Content-Based Image Retrieval（CBIR） system, a novel approach based on Pulse-Coupled Neural Network（PCNN） for texture image retrieval is proposed. PCNN is a new generation of artificial neural networks and powerful in data processing. The outputs of PCNN and Intersecting Cortical Model（ICM） represent unique features of original stimulus are invariant to translation, rotation, scaling and distortion, which is particularly suitable for content-based image retrieval system. By adopting PCNN and simplified PCNN model ICM, the dual value image sequence corresponding to different gray levels is obtained. The variance of each image in entropy sequence is then calculated to convert into one dimensional eigenvector used to represent the image features. The Euclidean distance is used to compute the similarity between images. A texture retrieval system based on query image is developed. The experimental results show, compared to the wavelet package transform, this approach can not only robust to the noises in images similarity retrieval and reduce the dimension of feature vectors, and has the property of shift, scale and rotation invariance, but also get higher accuracy rate.

Key words: Content-Based Image Retrieval（CBIR）, Pulse Coupled Neural Network（PCNN）, Intersecting Cortical Model（ICM）, feature extraction