Building image retrieval using color and line features

This paper proposes a building image retrieval algorithm using line features invariant to rotation as well as the color feature robust to illumination change. With a query image, the proposed algorithm consists of two steps: selecting candidate images among database images using the color feature and separating the nine best matching images from candidate images using the line features. In the first step, the proposed algorithm uses the hue histogram. We calculate the sum of absolute differences of hue histograms between the query image and database images, from which the similarity measure is computed for selecting candidates from database images. In the second step, a number of Hough peaks are found using Canny edge detector and the Hough transform. From the Hough array, the peak percentage and the distance ratio features are extracted. The circular correlations of them between the query image and database images are used to compute the similarity measure for ranking candidate images. Experimental results show that using the color feature gives better results than the line feature based method that does not use the color feature. The proposed building image retrieval algorithm improves the accuracy and reduces the processing time through cascading two steps using the color and line features.