Research Of Accuracy And Real Time Performance Of Auto Focusing System

Auto focusing is one of the most important technologies in digital image processing. Now many digital products, such as digital camera and digital video, are getting more popular. At the same time, some instruments, like microscope and scanner, are developing. Moreover, auto focusing has a widely use in GPS and computer vision. Based on those applications, auto focusing is being paid more attention to than several years ago.Following is the principle of auto focusing. Firstly, get images of different focusing level of the scene through image-capture instrument. Transforming the analog signal into digital signal, then send it to computer. Secondly, computer calculates the evaluation function values of these images and sequences those values according to mountain climbing searching algorithm. Thirdly, characteristic of focusing evaluation function illustrates that the image with the biggest function value has the highest focusing level. Equipments take and save that image.Through these analyses, auto focusing mainly has three modules, including focusing region selection module, focusing evaluation function module and mountain climbing algorithm module. Focusing region selection module decides which pixels take part in evaluation function. Actually, all pixels taking part in evaluation will be most accurate. However, if the size of image is huge, it is a waste of time that all the pixels take part in calculation. Only those important ones need to be considered. That is the main work of region selection module. Focusing evaluation function aims at evaluating the level of focusing, or defocusing in another word, thus to supply a foundation for controlling the lens. So it is ideal that focusing position's function value obviously distinguishes from other positions' values. Mountain climbing algorithm sequences the array of evaluation function values and finds out the biggest one. It directly influenced the speed and accuracy of auto focusing system.In the region selection module, traditional focusing region selection algorithm has a precondition that almost all the main objects locates at the image center. This precondition is not adaptive, especially when the main object deviates from image center. A new region selection method is proposed in this thesis, which bases on the first order of image. No matter where the main object locates at, focusing region can track it. This method enhances the covering rate of foreground object and increases focusing exactness.Also, a new focusing evaluation function with good anti-noise capability has been proposed. Firstly, the proposed function abstracts high frequency components through MRA (wavelet Multi-Resolution Analysis). Some neighbor correlation exists in each sub-band coefficients. But noise is random. So the noise coefficients have no such correlation. The new function introduces a frequency sub-band threshold, supposing that the component whose value is smaller than the threshold is noise. Filter it. In this way, noise and image signal could in general be separated.All the proposed algorithms have been simulated and tested by simulator and hardware platform. Results show many advantages of the new algorithms, such as high focusing speed, better accuracy and sensitivity, especially progressed in anti-noise capability.