Performance evaluation of image retrieval systems using shape feature based on wavelet transform

Digital era has produced large volume of images which created many challenges in computer science field to store, retrieve and manage images efficiently and effectively. Many techniques and algorithms have been proposed by different researcher to implement Content Based Image Retrieval (CBIR) systems. This paper discusses performance of different CBIR systems implemented using combined features colour, texture and shape as a prominent feature based on wavelet transform. Choice of the feature extraction technique used in image retrieval determines performance of CBIR systems. In this paper evaluation of performance of three CBIR systems based on wavelet decomposition using threshold, wavelet decomposition using morphology operators and wavelet decomposition using Local Binary Patterns (LBP) is done. Also the performance of these methods is compared with the existing methods SIMPLIcity and FIRM. Average precision is used to compare the performance of the implemented systems. Results indicate that performance of CBIR systems using wavelet decomposition give better results than simplicity and FIRM, also wavelet decomposition with Local Binary Patterns (LBP) exhibit better retrieval efficiency compared to wavelet decomposition using threshold and morphological operators. Theses CBIR systems have been tested on bench mark Wang's image database. Precision versus Recall graphs for each system shows the performance of respective systems.