Content-based Image Retrieval Speedup Based on Optimized Combination of Wavelet and Zernike Features Using Particle Swarm Optimization Algorithm

This paper presents a novel method to speedup content-based image retrieval (CBIR) systems. The proposed method can be very useful for retrieving images from a large database. For this task, Zernike and Wavelet features are first extracted from the query image, then an interval of potential matching images is computed from the database images using the extracted feature. Therefore, the query image is compared with images in the interval rather than the whole database which to speedup the retrieval process. Particle swarm optimization is employed to select relevant features among Zernike and Wavelet features, which leads to decrease feature extraction time. Three types of experiments are conducted to evaluate effectiveness the proposed method in terms of database reduction, retrieval accuracy and retrieval time. In the best case, the Corel-1k database is averagely reduced up to 33.98% from its original size, and preserving 71.92% of relevant images. Retrieval accuracy in reduced database is increased by 1% in comparison with retrieving from the original database. Meanwhile, the retrieval time is reduced up to 58.57% in comparison with retrieval time from the original database.

Keywords:

content , based Image Retrieval Particle Swarm Optimization Speedup Wavelet Features Zernike Moments