基于光场相机的子孔径图像提取和人脸检测应用

发布时间：2018-03-02 10:28

本文关键词： 子孔径图像像素重组 LBP算法傅里叶重聚焦方法人脸检测　出处：《太原科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着科技的发展,光场成像技术也扮演了越来越重要的角色。光场相机作为一种新型相机,也获得了越来越多的关注。传统相机拍摄的场景图像是只记录场景的二维信息,而光场相机则可以记录场景的所有光场信息,通过计算后可得到场景的多视点图像和不同聚焦面的像,甚至可以得到场景的全聚焦图像。因为光场相机的多视点图像获取技术是不开源的,而多视点图像的应用需求和前景非常巨大,所以本文提出了一种基于微透镜型光场相机的子孔径图像的提取方法;利用了光场相机在重聚焦方面的优势,结合LBP人脸检测算法,证明了光场相机拍摄的人脸照片可以较好地检测到在不同聚焦层的人脸。主要工作和创新点如下:1)提出了一种基于微透镜型光场相机的子孔径图像提取方法。该方法中,首先根据光场相机所采集到的光场数据,其主透镜下的一个子孔径在CCD上的成像与光场相机中所有微透镜单元成像区域具有相同的坐标,光场相机所得到的子孔径图像就相当于光场相机主镜头光圈减小后生成的像。同理,该像也等同于微透镜阵列的等效像元阵列所成的像。所以我们要先利用峰值检测法标定出各微透镜图像的中心点;之后提取以中心像素点为半径的一块明亮区域,将区域内相同位置的像素点全部提出;最后进行重新排列拼成一副完整的具有像素级别的微小视差图像,该图像即为所需的子孔径图像。2)提出了一种基于光场相机重聚焦功能的人脸检测应用。该方案中,在结合光场相机数字对焦和数字重聚焦的原理下,利用LBP人脸识别算法在纹理清晰的照片上的独到优势。先利用光场相机的调焦特性得到许多同一场景下不同焦距的照片;之后利用傅里叶切片定理将图像中所有人脸全部进行重聚焦,将得到的不同深度重聚焦图进行图像融合后得到一副完整的全聚焦图像;最后和单一焦距层的照片进行人脸检测结果对比,结果证明了经过光场相机全聚焦后的图片可以在不同深度上都检测到人脸,而单一聚焦层的照片却只能在聚焦点附近深度才能检测到人脸,体现了光场相机在人脸检测方面的潜力和优势。
[Abstract]:With the development of science and technology, optical field imaging technology also plays an increasingly important role. As a new type of camera, optical field camera has attracted more and more attention. The light field camera can record all the light field information of the scene, and the multi-view image of the scene and the image of different focusing plane can be obtained by calculation. The full focus image of the scene can even be obtained, because the multi-view image acquisition technology of the light field camera is not open source, and the application demand and prospect of the multi-view image is very great. Therefore, this paper proposes a subaperture image extraction method based on microlens light field camera, and combines the advantages of light field camera in refocusing and LBP face detection algorithm. It is proved that the face images taken by the light field camera can detect the faces in different focusing layers. The main work and innovation are as follows: 1) A subaperture image extraction method based on the microlens light field camera is proposed. Firstly, according to the light field data collected by the light field camera, the imaging of a sub-aperture under the main lens on the CCD has the same coordinates as all the imaging regions of the microlens unit in the light field camera. The sub-aperture image obtained by the light field camera is equivalent to the image generated after the aperture of the main lens of the light field camera is reduced. This image is also equivalent to the image of the equivalent pixel array of the microlens array. So we first calibrate the center points of each microlens image by using the peak detection method, and then extract a bright region with the radius of the central pixel. All pixels in the same position in the region are proposed. Finally, a complete set of tiny parallax images with pixel level is rearranged. A face detection application based on the refocusing function of a light field camera is proposed. In this scheme, the principle of digital focusing and digital refocusing of the optical field camera is combined. The unique advantage of LBP face recognition algorithm in textured images is presented. Firstly, many images with different focal lengths in the same scene are obtained by using the focusing characteristics of the light field camera. Then all the faces in the image are refocused by Fourier slicing theorem, and a complete set of full focus image is obtained by fusion of the different depth refocusing images. Finally, the results of face detection are compared with those of single focal layer. The results show that the images with full focus of the light field camera can detect the face at different depths. But the single focus layer can only detect the human face in the depth near the focal point, which shows the potential and advantage of the light field camera in face detection.
【学位授予单位】：太原科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】