A new approach for detecting local features

Local features up to now are often mentioned in the meaning of interest points. A patch around each point is formed to compute descriptors or feature vectors. Therefore, in order to satisfy different invariant imaging conditions such as scales and viewpoints, an input image is often represented in a scale-space, i.e. size of patches are defined by their corresponding scales. Our proposed technique for detecting local features is different, where no scale-space is required, by dividing the given image into a number of triangles with sizes dependent on the content of the image at the location of each triangle. In this paper, we demonstrate that the triangular representation of images provide invariant features of the image. Experiments using these features show higher retrieval performance over existing methods.