Studies On The Theory And Implementation Method Of Digital Autofocus Technology

As a key technology in digital imaging system, autofocus plays an important role in terms of improving performance of digital imaging products. In order to solve current problems in autofocus field, this dissertation fulfilled some important theoretical studies, such as autofocus theories and methods, focus measures, focus algorithms and focus framing technologies. Furthermore, we have designed and construct two autofocus systems based on PC and DSP respectively. Funded by the Financial Aids for Cadre College Teachers of the Education Ministry, this dissertation achieved a numbers of outcomes.Firstly, basic theories and methods of autofocus based on image processing are studied in detail. In the method DFF (depth from focus), aiming at the characteristic of the images, we proposed that Wiener filter and image brightness normalization can be apply as preprocessing tools. As a result, smoothness and acuity of focus measure curve have been improved remarkably. In the method DFD (depth from defocus), feasibilities of restoring images from defocus images and estimating the focused position are discussed and confirmed. We acquired best restoration of defocus images using one blurred image, and estimated the focused position according to radius of blurred circles at two different lens positions, the focus error was less than 5%. Furthermore, a new DFD method was put forward. The method estimated the focused field through analyzing three images at different lens positions and then found best imaging position within the field in small steps. Our method has some advantages of rapid reaction, high accuracy and good performance for the objects with edge features.Secondly, focus measures and focus algorithms are studied. For focus measures, frequency spectrum functions, gradient functions and entropy function are analyzed in detail. We revised frequency spectrum functions by using weighting factor - the distance between direct component and alternating component, and improved the acuity of focus measure curve. In addition, the dissertation studied color interpolation algorithms of imaging sensor, and proposed two new approaches. The first one was based on green components, the second one was taken color filter array (CFA) into account. Keeping acuity of focus measure curve, new approaches reduced run-time and improved the speed of focus.Thirdly, intelligent focus framing technologies are studied. Appropriate focus window selection is an important factor for focus. We probed into four schemes. Statistical tools were applied to analyzing possible fields of imaging objects, and possibilities of wrong focus were reduced greatly. Variational image sampling was employed for focus windows selection, andthe harmony between resolution and field of view was implemented. A human face extracting method based on color, shape and detail features was proposed, in order to overcome restriction of method based on skin detection. A method using optical flow was introduced to segmenting moving human face in focus images, and has better perspective in surveillance situation. The feasibility of above schemes was proved by large numbers of experiments.Then, two autofocus systems based on PC and DSP respectively have been designed and construct. The systems perform the autofocus action with the following steps: analyzing image blurred degree, moving motorized lens, finding best imaging position and finish autofocus. A series of focus experiments have been performed on the above two autofocus systems, their veracity and reliability have probe to be excellent.Finally, main conclusions of this dissertation have been summarized, and the prospects and study emphases of further work have been forecast.