图像局部模糊检测的SVM方法研究

发布时间：2018-04-21 23:39

本文选题：局部模糊图像 + 模糊特征　；参考：《武汉大学》2017年硕士论文

【摘要】：图像在信息的表达与传递中有着关键的作用,人们最常用来记录图像信息的工具就是相机。当相机拍摄图像时不可避免会存在图像失真的情况,比如模糊、色偏等。而图像退化后会大大地降低图像信息表达的精准度。本文首先分析了运动模糊与失焦模糊两种局部模糊的形成机制。然后对比两种局部模糊图像的退化机理,寻找两者的特点与区别。从研究运动模糊与失焦模糊图像的形成机理中,也可同时发现模糊与清晰图像间的差别。由于局部模糊图像的模糊核不易精准地计算,所以本文未采用估计图像模糊核的方法来区分模糊的类型。而是提出了四项模糊特征用于区分模糊与清晰,并判断图像的模糊类型。提取多个模糊特征作为分类的特征数据,是为了更好地从各方面表示图像的信息。在图像库中,对所有的图像块采用SVM进行学习与训练,建立图像类型的预测模型。利用两次SVM分类器可将图像分类为清晰图像、运动模糊图像与失焦模糊图像三类。最后,利用上述建立的SVM分类模型,对局部模糊图像以每个像素为中心分块进行检测,判断其是否为模糊,完成图像局部模糊区域的检测。由于图像检测存在误判现象,所以最终检测得到的二值化效果图中会出现小空洞或不连接等现象。最后,采用数学形态学算法对最终的试验效果进行完善。图像局部模糊分类试验结果证明了本文提出的模糊特征能够较为准确地判断图像块是否为模糊图像,并且能够辨别图像块的局部模糊类型。这也说明了将该SVM分类模型应用于图像局部模糊区域检测有一定的可行性与可靠性。在局部模糊检测试验中,最终的检测效果图证明了本文提出的图像局部模糊检测的SVM方法具有较高的准确率。对比于理想状态下的局部模糊检测效果图,该局部模糊检测算法能够较为准确的检测到一幅局部模糊图像中具体的模糊区域。
[Abstract]:Image plays a key role in the expression and transmission of information. The most commonly used tool for recording image information is the camera. When the image is photographed, the image distortion is unavoidable, such as blurred, color deviation, etc. and the image degradation will greatly reduce the accuracy of the image information expression. This paper first analyzes the transport of image information. Two local fuzzy formation mechanism of dynamic fuzzy and defocus fuzzy. Then the characteristics and differences of the two local fuzzy images are compared. The difference between the fuzzy and the clear images can be found at the same time from the study of the formation mechanism of the blurred image of the motion blur and the defocus. The method of estimating the fuzzy kernel of the image is not used to distinguish the fuzzy type, but four fuzzy features are used to distinguish fuzzy and clear, and to judge the fuzzy type of the image. Multiple fuzzy features are extracted as the feature data of the classification. It is to better express the information of the image from all aspects. All image blocks are studied and trained by SVM, and the prediction model of image type is established. Two SVM classifiers can be used to classify images into clear images, motion blurred images and blur blurred images. Finally, using the SVM classification model established above, the local blurred image is examined with each pixel as the central block. Determine whether it is fuzzy to complete the detection of the local blurred region of the image. Because of the phenomenon of misjudgement in the image detection, there will be a small hole or no connection in the final two valued effect map. Finally, the result of the final test is perfected by mathematical morphology algorithm. The local fuzzy classification test knot of image is used. The results show that the fuzzy feature proposed in this paper can accurately determine whether the image block is a fuzzy image, and can distinguish the local fuzzy type of the image block. It also shows that the application of the SVM classification model to the local fuzzy region detection is feasible and reliable. In the local fuzzy test, the final detection is tested. The test results show that the SVM method of local blurred image detection proposed in this paper has high accuracy. Compared with the local fuzzy detection effect map under ideal state, the local fuzzy detection algorithm can detect the specific fuzzy region in a local blurred image more accurately.

【学位授予单位】：武汉大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】