An Improved SURF Algorithm with FREAK Descriptor for BGA Component Image Stitching

Abstract

This paper presents an image stitching method of BGA component for industrial operation. The mounter is the core equipment of the surface mount system, and it currently has no way to place components larger than the camera's field of view. Therefore, a high-speed image stitching method of BGA component is still not available. In this method, the hu invariant moment is used to determine the overlapping area of the image and reduce the range of feature point extraction. After obtaining the feature points, the differences between the SIFT (Scale Invariant Features Transform) algorithm, the SURF (Speeded Up Robust Features) algorithm and the improved SURF algorithm are compared. The improved SURF algorithm combines the SURF feature extraction method and the FREAK (Fast Retina Keypoint) descriptor is developed. After the feature points are processed by BFMather matching and RANSAC algorithm, the image transformation matrix of the two images is calculated. Finally, the perspective transformation is performed on the image to complete the image stitching. This image stitching method is shown to be simple and effective for high-speed industrial application. The procedure taken can be extended to other similar IC component and to other machine.