A new fast and accurate training-based auto-focus for mobile-phone cameras

In this paper, we propose a fast and accurate auto-focusing method for mobile phone cameras. The proposed method first extracts a feature vector consisting of the focus value increment ratios at the predefined lens positions. Then, the best in-focus lens position is estimated by comparing the input feature vector and the training database. Experimental results show that the proposed method is superior to the conventional methods in terms of the searching speed.