数码相机中CCD/CMOS图像传感器的饱和电流和背景噪音使得一次曝光很难超过60-80dB的动态范围。根据人眼对灰度图像的识别特性，可以采用对相同景物二次不同曝光图像软件合成的算法来获取高动态范围的图像。实验结果显示，当景物画面各部分光照明显差别时，这种方法能有效地获得全部细节信息。

In avigation, navigation and other fields, electrical image stabilization technique (EIS) is applied widely. Recent research for electrical image stabilization is focused on motion vector estimation, which influences accuracy of image stabilization and its applicable scope directly, and pay less attention to the relation between image stabilization algorithm and structure character of image sensor. In fact, while image sensors are interline transfer CCD, 2D rigid motion model is not accurate enough for the image stabilization algorithm. There exist the displacement between the rows in a frame, for the different rows are exposed at different time. In order to compensate the interline displacement caused by the movement of the sensor both in horizontal and vertical direction, an image revision method based on correlation was put forward for the structure character of interline transfer CCD. During exposure of each frame, the motion of camera is along uniform direction and approximately even (otherwise it is beyond application scope of EIS), so it is feasible to revise image according to the displacement at the top and the bottom of image. Image sequence was acquired by telescope with function of optical image stabilization, and then was processed by electrical image stabilization algorithm. Experiment results showed that the image fidelity has been improved remarkably after image revision processing.

运用生物神经系统的神经信息处理方式开展对自动人脸识别的研究。采用主要成分分析技术—Karhunen—Loeve变换，构造了一个能够将图像空间的人脸映射到人脸空间中，并实施识别的实验系统。分析7该系统的构成与特点，并给出了实验测试结果。

CCD/CMOS图像传感器的饱和电流和背景噪音使得一次曝光很难超过60-80dB的动态范围。用LIP（Logarithmic image processing）模式将图像转换为灰色调函数，再将基于LIP模式的非线性多重微区域图像分解算法进行改进，可以适应不同亮度的单幅图像处理，提高了可分辨信息量和图像对比度，同时抑止了噪声。质量评价显示，该算法能将MOS 等级四的源图像提高到等级一。

Using pattern recognition to select focus windows, the primary imaging object s which are men (PIOM for short) is discussed in details. The necessity of utilizing pattern recognition in auto2 focus is illust rated. The skin2color dist ribution characteristics are used to detect skin regions in PIOM images. Morphological operations are employed in the process to get ride of interspaces arising f rom eyes, nost rils, mouths and other non2skin part s among skin areas. Then the detected skin regions are chosen as the focus windows. Experimental result s show the realistic applicability and efficiency of this approach.