|
This paper presents a method to compare an image with a set of images and rank the images by their similarity. The common way either uses time-cost features or a set of training examples to get arguments, this method is fast and similar with text comparison. Firstly, we get all images' corner points by chord-to-point distance accumulation (CPDA), then generate their feature vectors by corners which can overcome different lightness, rotation and scale, finally compute the similarity between the images by their feature vectors, while the similarity function provides two arguments to fit into different applications. The experimental results tell us that the precision and performance both perform very well. As this method is kind of like text comparison, so it is well-defined for the image search engine. |
|
Keywords:image similarity; corner detection; image search engine |
|