Autofocusing In Low Signal-Noise-Ratio Environment

Autofocus (AF) is a key technique in digital image capture system to help the imaging instruments focus on the target objects automatically. The demand for auto-focusing techniques is gradually increasing in many visual applications such as digital cameras, camcorders, video surveillance systems and microscopes.An AF system involves three main elements: a focusing window which defines the region of the scene that is to be focused, a sharpness function that evaluates the image sharpness, and a searching strategy to find the global maximum of the sharpness function. Currently, most imaging systems adopt this semi-digital auto-focusing technique, which has the analysis module to determine the focusing status by computing the sharpness of the input image, and the control module to move the focusing lens back and forth until the optimally focused image is obtained.A variety of window selecting methods and image sharpness functions and peak searching strategies have been proposed in the literature. However, images captured in low signal-to-noise ratio (SNR) environments usually suffer from a lack of sharpness due to the failure of the camera's passive auto-focus system to locate the peak in-focus position of a sharpness function that is extracted from the image. In conditions of low light, low contrast or complicated background, the sharpness function becomes flat, making it quite difficult to locate the peak.In my work, a systematic approach will be introduced to address the problem of low SNR AF by performing computationally simple image enhancement preprocessing steps and improving the stability and robustness of the AF algorithms. Firstly, a large amount of image samples must be taken to embody a variety of information and simulate the ambient environments. From these experimental materials, a relation model between various imaging parameters and AF performance will be founded. Secondly, fine image enhancement and restoration methods must be adopted to ensure that images are clear enough for analysis. Thirdly, a good AF system with an optimal focusing window selecting method is necessary. The familiar selecting approaches are central region method and multi-region method. For the purpose of capturing the main body of the image, the multi-region method has the relative superiority. But as a result of utilizing the weighted multi-region average approach to calculate the sharpness of the image, the variation range of the image sharpness evaluation is reduced so that the accuracy of AF is influenced to a certain extent. a Gaussian non-uniform sampling method has been presented, but it also depends on the assumption that the interesting object is near the center of the image. Other approaches, such as face detection and skin detection, are limited to the applications for portraits photographing and are time consuming as well. Because of the application limitations of the existing focusing region selection methods and the wide variety of imaging conditions, a focusing window that may be dynamically selected or changed according to images is required. In my work, some criteria for assessing different focusing windows will be presented by analysing the contents of the images. Then, to comply and optimize those criteria, a novel focusing window selection method based on edge masking will be discussed. Finally, new algorithms and approaches will be tested on our auto-focus verifying platform. In conclusion, the bottleneck and improvement advice on auto-focus algorithm will be given.