Research on the Face Image Detection in Coal Mine Environment

Transcription

1 2016 International Conference on Electronic Information Technology and Intellectualization (ICEITI 2016) ISBN: Research on the Face Image Detection in Coal Mine Environment Xiucai Guo and Mi Liu ABSTRACT In recent years, face recognition system used in the coal mine workers identity becomes a hot spot. face detection is the key to face recognition, but at the special coal mine working environment, the monitoring image contrast is low, the light intensity of illumination is weak. The miners face image detection becomes difficult. In order to solve this problem, The light density strengthen based on homomorphic filtering, the image noise removal based on wavelet transform was put forward. Canny edge detection method combining the extraction of face and mine cap contour method was used to detect human face area. The experiment shows that this method is well implemented in face detection and is helpful for further identification. INTRODUCTION Coal is our country the main one-time energy. Because of the complexity and the extensive number of the coal mine workers and the hostile environment of coal mine. Coal mine personnel management has many problems[1]. the way of radio frequency identification technology is one of the most common ways to check on work attendance[2]. But this method has the problem of lack of authentication. Both at home and abroad for face recognition application research on coal mine safety management system is started in recent years. for example, in literature [2-3]. The application in coal mine is limited to the well personnel attendance so high efficient personnel identification is very important, face Xiucai Guo, Mi Liu, Xi an University of Technology, Shanxi, China,

2 detection is the key to face recognition, therefore the research of the coal mine face detection is of great significance[4]. Image Preprocessing Method Image preprocessing methods include image histogram equalization, gray stretch method, median filter and homomorphic filter and so on. Under the coal mine, as the specific location of the illumination condition, Monitoring exists the problem of low distinguishability. Under different illumination conditions, facial image homomorphic filter pretreatment method can be used to avoid light to its influence. Homomorphic filtering processing method will gray level transformation and frequency filtering combine the two methods. Using the method of homomorphic filtering is to change the original image of the light intensity and the characteristics of the reflected light intensity, It not only can reduce the dynamic range of images, but also can increase the contrast. Homomorphic filter pretreatment of face recognition image. The experimental results are shown in Fig1. Comparing with the original image, the details of the processed image are more clear, contrast is enhanced, the image effect effectively improves. Wavelet transform for image compression and finally the image is decomposed into half length of low frequency and high frequency components, the low frequency component of frequency subdivision, high frequency component for the time segment[5]. After the wavelet transform, the useless information of face image is compressed, facial features embodied completely. the method of using wavelet transform to reduce the dimensions of the image. Single scaling wavelet transform to the experimental results are shown in Fig 2. After wavelet transform, the image is broken into four parts, respectively is: the approximate image, the level of detail, vertical and diagonal details. High compression ratio, compression after can keep the characteristics of the image remains the same. By a wavelet transform, a 448 *600 image can be reduced to 224*300, dimension is reduced to a quarter of the original. The image is multi-scale refinement and fully highlight its characteristics. 376

3 The original figure grayscale spectrum image before homomorphic filtering spectrum image after homomorphic filtering image after homomorphic filtering Figure1. the original image of homomorphic filtering and spectrum image contrast diagram. (note: the images from sina) Face Detection Realization Face detection technology in accordance with the general image can be divided into two kinds: one kind is based on static image, the other based on video image which detects the face image in the video detection, if it contains face region will mark in the region. The latter one is more difficult than the former. Because it have higher real-time request for algorithm [6]. the former is mainly divided into three categories: face feature recognition method, her face recognition and other space analysis methods [7]. Face detection method based on color images, The color images can provide abundant information, thus images can make use of their own characteristics. Under the coal mine, people must wear a helmet. according to the characteristic, the helmet face contour model can quickly detect the human face area. the detection is divided into two pieces, the detection of helmet part and the detection of face. The sample picture is a 448 * 600 image, as shown in Fig 3. rectangle ABCD is as mine cap area, the left Angle A as pixels origin, coordinates of point A, B, C, D, are (10,120), (10,340), (140,120), (140,340).We can work out the area of the rectangle. It is 220 * 130 = pixels. Rectangular CDEF as the face range point E, F coordinates, respectively (280,120), (280,340), The face area is 220 * 140 =30800 pixels. In the same image, It s visible that the helmet area is similar to the area of the people and the helmet width greater than the face width. Of course, the proportion of face can also be work out. After adjusting the canny threshold, we can make the main parts of the image. mine cap and facial 377

4 contour is outstanding. When the threshold value is 74, the edge effect is best, after repairs the edge crack, results are shown in Fig 4. Figure 2. wavelet transform to each detail images. Figure 3. face image pixels proportion. Mine cap at the top and use the mine cap model detection, use a rectangle to approach cap edge area, and then according to the characteristics and the facial contour area and the distance from mine cap, determines the face region and marked in the artwork, uses the same identification box to select the face area, the result is shown in Fig 5. CONCLUSION After analyzing the particularity of mine environment and comparing with various algorithms combined with the advantages and disadvantages of each method, according to the face and mine hat features joint detection face region, through the experiment, The method accurately detects the mine face. It lays a foundation for the rest of the face recognition and has strong and high practicability. 378

5 face detection area result figure Figure 4. edge detection effect image. Figure 5. The face detection result. REFERENCES: 1. Guang Yang, Qingbin Yuan, Qiang Zhao, Weicheng Zhou Face recognition in the coal mine environment review. The 24th Session of the National Coal Mine Automation and Informatization of the Academic and the 6th Chinese Coal Mines Informatization and Automation Top BBS In Chinese. 2. Fulin Chun, Sheng-Fuu Lin Efficient face detection method with eye region judgment. Eurasip Journal on Image and Video Pressing, 2013: Llichao Chen, Xiuqin Zhang, Lihu Pan, Bo Li Coal mining face recognition algorithm in the system of check on work attendance study. Journal of Industrial Automation, v. 41: No : In Chinese. 4. Hongdong Liu, Shaoyi Jia Integrated coal mine personnel management system application technology. Coal Science and Technology In Chinese. 5. Yumei Niu, Lichao Chen, Lihu Pan The coal mine workers face contour extraction method [J]. Journal of Industrial Automation. 9 (1): In Chinese. 6. Huijun Shen Research on facial recognition image preprocessing method and implementation and innovation. Science and Technology. (18): In Chinese. 7. Xing Huang, Xiaotao Wang, Lihua Liu A fast and efficient face detection method. Computer Engineering and Application In Chinese. 379