Hough transform for rotation invariant matching of line-drawing images

Hough transform can be used for indexing of line-drawing images for content-based image retrieval. Angular information is used for generating the feature vector (index) as it gives global description of the image, allows compact indexing, fast retrieval and scale, translation and rotation invariant matching. In the case of very large images, however, the angular information is not always sufficient to differentiate images from each other. To alleviate this problem, we extend the idea by including also positional information of the lines in the feature vector. This gives more representative description of the images and therefore allows more accurate image matching. The main problems of this approach are: (1) to keep the feature vector compact, and (2) to preserve the property of the matching being translation and rotation invariant. We give solutions to both of these problems and introduce a new indexing scheme, which has better matching accuracy but at the cost of slower retrieval time