Autofocus is critical for a camera system due to its massive impact on image quality. It is essential to get correct focus on the region of interest without user interaction. Contrast based focusing is the most prevalent form of focusing which uses statistics from the image signal processor (ISP) to guide lens movements. An autofocus system consists of numerous hardware and software components and each component is especially optimized. Autofocus system design consists of repetitive and tedious field tests on real scenes. This approach, however, is very time consuming and laborious.This thesis presents an optimization methodology to expedite autofocus design and improve camera performance. We propose the use of genetic algorithms (a branch of evolutionary algorithms) to improve autofocus. Genetic algorithms are derived from the biological model of evolution and natural selection. In this thesis, we create an environment in which potential solutions can evolve. We demonstrate the effectiveness of our method by optimizing focus kernel and step-length of a camera lens. The fitness of these parameters is measured using the contrast from focus statistics, shape of the focus statistics curve and time taken for focusing.Input images from different ambient conditions are captured using consumer phones and are utilized in optimization to get effective clues for autofocus. Autofocus design resulting from our methodology is tested in retail phones to verify its application and efficiency. Improvements in autofocus are observed including enhanced contrast extraction and reduced time-to-focus. Biggest performance gain is seen in low-light as the available contrast is low and it becomes even more important to obtain better focus statistics. Our autofocus design results are deployed to commercial camera phones which proves the effectiveness. Optimizing camera autofocus is a very industry specific topic and this thesis presents a possible solution to this optimization problem.
展开▼
