Scalable Image Search Re-ranking through Content Based Image Retrieval (CBIR) Method

There are number of ways for searching images through popular search engines like Google, Yahoo, Bing etc..,. The latest survey in the Data Mining proves that there is a vast increase in the percentage, in searching the images related to the text, while surfing the Internet for the personal, educational and professional as well. The perpetual and former ways for searching the images from the net and Re-ranking endure from the unreliable ranking assumptions than the initial text based image search results that are used within the remote Re-ranking methods. This paper is designed mainly to focus and to propose a prototype based Re-ranking technique to deal the drawback of text based image searching in a scalable fashion. Validation aspects like Energy, Entropy, Contrast, Homogeneity, shape, color, skew and Euclidean distance measures are considered. K-Means clustering, SVM classifier and Re-ranking algorithmic technique are used to get the efficient prototype based scalable result. The experimental results on a representative internet image search dataset comprising of 350 queries demonstrate that the projected technique outperforms the present supervised and unsupervised Re-ranking approaches. Moreover, it improves the performance and precision over the text based image searching by twenty-five percent.

Download PDF