Recognize the Image of the Inscription at the Bottom of Purple Clay Teapot Using Improved Sift Algorithm

Abstract

We use an improvement to the conventional sift algorithm to accurately recognize the image of the inscription at the bottom of purple clay teapot. We show that the improvement causes the false match rate of the image to be greatly reduced, and that the improvement meets the requirement of high-precision project such as purple clay teapot. We propose a method of adaptively selecting stable points. The feature points extracted in this way have smaller changes and are more stable with scale changes, which is more conducive to image matching. We extensively test our algorithm using a dataset of one hundred sets of images.