A Novel Training Based Auto-Focus for Mobile-Phone Cameras

Abstract

In this paper, we propose a fast and accurate training based auto-focus (AF) method for mobile-phone cameras. Given a set of training data, the proposed method collects feature vectors consisting of focus value increment ratio. The representative feature vectors corresponding to every possible best in-focus lens position (BILP) are stored in the database. In the proposed training based AF method, the BILP is obtained by comparing the input feature vector with the representative feature vectors in the database. To further enhance the accuracy of the proposed AF method, we also introduce a new focusing window that is effective in detecting the target object. The experimental results show that the proposed method can accurately estimate the BILP faster than the existing methods.