一种用于小人脸精准检测的图像超分辨率算法

An Image Super-Resolution Algorithm for Accurate Detection of Small Faces

针对实际应用中视频流的小人脸检测难度大、误检率高的问题,结合人脸图像结构信息提取,提出基于图像超分辨率的小人脸检测模型,设计一种应用于小人脸精准检测的改进POCS图像重建算法,该算法首先对小人脸图像进行图像增强(直方图均衡化、图像降噪、边缘增强),以获取小人脸结构信息,再将增强后的图像作为POCS超分辨率算法的基础帧,并在传统POCS算法中利用三次样条插值和中值滤波进行边缘优化超分辨率重建,并采用最小绝对差分配准则的块匹配算法进行图像配准和图像矫正,最后将三次插值、基本POCS小人脸图像以及本文提出的改进超分辨率图像输入到Dlib人脸检测模型中进行测试,实验结果表明本文模型的可行性和所提出的超分辨算法的有效性。

In view of the difficulty and high false detection rate of small face and multi face detection in video stream in practical application,com⁃bined with the characteristics of face image structure information and image super-resolution recovery algorithm,this chapter proposes a small human face detection model based on image super-resolution,and designs an improved POCS image super-resolution reconstruction algorithm for small face accurate detection Small face image is processed by image enhancement to obtain the structure information of each small face.On this basis,the enhanced image is taken as the basic frame of POCS super-resolution algorithm.In the traditional POCS algo⁃rithm,cubic spline interpolation and median filter are used to optimize the edge super-resolution reconstruction,and the minimum abso⁃lute is used Finally,the original small face image,the basic POCS small face image and the improved super-resolution image proposed in this paper are input into Dlib face detection model for testing.The experimental results show the feasibility of the proposed model and the effectiveness of the proposed super-resolution algorithm.