【作者】 周杰；

【导师】 赵群飞；

【作者基本信息】 上海交通大学 ， 机械电子工程， 2007， 硕士

【摘要】 现在,数字成像技术日新月异,但其基本原理相对以前并未发生根本性的改变,比如曝光方法。曝光作为成像设备最重要的成像因素之一,是衡量数码图像质量的主要指标。目前绝大多数数码相机都具备相当的自动曝光功能,其原理就是借助自动测光系统获得数码图像的最佳曝光依据,再自动配置光圈及快门来进行拍摄。一般情况下,拍摄者只需简单的按下快门即可获得质量不错的图像。然而,自动曝光并非每次都能达到最佳的曝光效果,原因就是测光系统不能完全适应千变万化的光照环境,对于普通使用者而言,在拍摄过程中,如何通过相机自动曝光系统得到一个恰当的曝光值始终是个难题。因此,针对上述问题,本文首先从人脸检测等算法方面进行改进,根据人在照片中的位置来划分测光区域,以提高其智能性;其次,根据曝光与照片灰度直方图之间存在的对应关系,对各个区域进行数学统计,节省时间及内存;最后,引入模糊逻辑算法,根据已提取到的图像信息及专家经验,对不同的区域给出不同的曝光权重,得到最终的曝光量。本论文的研究内容主要包括三个方面:⑴设计适应性强,满足实时性要求的人脸检测算法。根据实际应用的要求,本文采用改进的AdaBoost方法和类Haar技术快速确定人脸区域,通过彩色补偿和光线补偿后使用肤色验证方法,确定人脸的最终位置。此后根据人脸与身体的比例关系,大致找出人体区域,并以此为基准划分图像区域。⑵根据图像曝光原理及其在对应灰度直方图上的表现,设计一种数学统计方法,通过它能够较直观的反映出图像的曝光状况,并且要求能在图像各个区域分别进行统计,速度快、资源占用低。⑶引入模糊逻辑算法,根据已经获得图像亮度信息,结合摄影学知识及自动曝光理论,设计模糊逻辑规则,最后计算得到图像不同区域的权重值。大量实验数据及对比表明,本文算法通过基于人脸识别的多区域分割、直方图统计及模糊逻辑系统对数码图像的曝光情况进行更好的分析和判断。对于那些较为常用的人像摄影及复杂明暗条件下的曝光判别都具有较好的适应性,同时还能够对同一场景下不同曝光补偿值拍摄的图像进行比较,其实现过程也相当简单,只需用户变换曝光补偿值再进行一次上述判断即可,经过与有经验的摄影师的肉眼评判结果比较,发现对于那些构图复杂或者奇异曝光的图像大大提高了判断的准确性和稳定性,且速度快捷,操作简单。更多还原

【Abstract】 Digital imaging technology has made never-ending changes and improvements these years, but the fundamental principle of which does not change a lot, such as the exposure method. As one of the most important imaging elements, exposure effect is the chief mark to evaluate the quality of digital images.Most of the recent digital cameras are equipped with auto-exposure function whose principle is to get the optimal exposure quantity through photometric system and adjust the aperture and shutter automatically. In the ordinary course of events, users can get a good image just by pressing the shutter button. However, it’s not so easy for auto-exposure system to get the best exposure effect every time. The reason is the photometric system can not suit complicated photo environment. It’s always a big problem to select a suitable exposure value for ordinary users. To solve this problem, the algorithm mentioned in this thesis first improves the existing face detection method, dividing the image into 6 regions according to the position of human beings; secondly, makes statistics based on the relationship between exposure effect and histogram; finally, pulls in fuzzy logic algorithm to give weights to different regions and determine the exposure quantity by image information and experts’experience. The main task of this thesis lies in the following four aspects:(1) Fast, accurate face detection & tracking algorithm. A core classifier based on Haar-like feature and AdaBoost learning algorithm. Before face detection, there are some pretreatments including ray and color compensation. After that, use skin model to validate the face area and ensure face location. Find human’s body according to the proportion of the width of head to that of shoulder. Divide the image with this conclusion.(2) To show the image’s exposure condition, design a statistics method based on the relationship between image’s exposure effect and its histogram. It also can be applied in different areas of the image and takes little time and resource.(3) Introduced the fuzzy logic algorithm. With image information and experts’experience, find input and output, design the fuzzy logic rules base and the fuzzy inference engine. Input the output quantity of fuzzy system into the weights formula更多还原