Multiscale Texture Analysis and Color Coherence Vector Based Feature Descriptor for Multispectral Image Retrieval

摘要

Content Based Image Retrieval (CBIR) for remote sensing image data is a tedious process due to high resolution and complexity of image interpretation. Development of feature extraction technique is a major portion to represent the image content in an optimal way. In this paper, we propose a feature descriptor which combines the color coherent pixel information and GLCM texture features in multi scale domain. Curvelet transform is used to decompose the image into coarse and detail coefficients. Then Gabor magnitude is computed for each coefficient to improve the directional information. GLCM texture features are extracted from the Gabor magnitude response. The novel feature set by combining the CCV and GLCM using curvelet and Gabor filter is developed. Mahalanobis distance measure is used to find the similarity between query and feature database. Average Normalized Modified Retrieval Rate (ANMRR) is computed to evaluate the performance with the state of art methods.