Content-based matching of line-drawing images using the Hough transform

Abstract.

We introduce two novel methods for content-based matching of line-drawing images. The methods are based on the Hough transform (HT), which is used to extract global line features in an image. The parameter space of the HT is first thresholded in order to preserve only the most significant values. In the first method, a feature vector is constructed by summing up the significant coefficients in each column of the accumulator matrix. In this way, only the angular information is used. This approach enables simple implementation of scale, translation, and rotation invariant matching. The second variant also includes positional information of the lines and gives a more representative description of the images. Therefore, it achieves more accurate image matching at the cost of more running time.