A Scale and Rotational Invariant Key-point Detector based on Sparse Coding

Most popular hand-crafted key-point detectors such as Harris corner, SIFT, SURF aim to detect corners, blobs, junctions, or other human-defined structures in images. Though being robust with some geometric transformations, unintended scenarios non-uniform lighting variations could significantly degrade their performance. Hence, a new detector that is flexible context change and simultaneously both illumination very desirable. In this article, we propose solution challenging problem by incorporating Scale Rotation Invariant design (named SRI-SCK) into recently developed Sparse Coding based Key-point (SCK). The SCK different fully invariant affine intensity change, yet it not designed handle images drastic scale rotation changes. SRI-SCK, the invariance implemented an image pyramid technique, while realized combining multiple rotated versions of dictionary used sparse coding step SCK. Techniques for calculation key-points’ characteristic scales sub-pixel accuracy positions are also proposed. Experimental results on three public datasets demonstrate high repeatability matching score achieved.