The hottest research on face detection based on ma

Research on face detection based on mathematical morphology

o introduction

computer face recognition is a cutting-edge topic in the field of pattern recognition and artificial intelligence, which has a very broad application prospect. Automatic face recognition system mainly includes face detection and location, and face feature extraction and recognition

mathematical morphology is a nonlinear theory based on set theory, which is the application of lattice theory in spatial structure. Its basic idea is based on the logical relationship between pixels rather than algebraic relationship. This kind of lamination of high-temperature wire and cable insulation is conducive to the geometric description of the image. Unlike other linear image processing theories based on algebra, it has the characteristics of not blurring the image boundary and details. It is a new theory and method for image processing

1 skin color region detection

generally, the image is stored in RGB, but the RGB representation method is not suitable for skin models. In RGB space, the three primary colors (R, G, b) represent not only color, but also brightness. Due to the change of ambient light, the brightness may make face detection more complex, which is not reliable in the process of skin segmentation. In order to make use of the clustering of skin color in the chromaticity space, it is necessary to separate the chromaticity information from the brightness information in the color expression. This can be achieved by converting RGB into a color expression space that separates chromaticity and brightness. In the experiment, YCbCr space was selected for skin color region detection. The CB and Cr components represent the chromaticity of blue and red respectively, which are stable when the illumination changes, and Y represents the brightness information that is easy to be changed. The color space can be obtained from the linear transformation of RGB format, and the conversion formula is

the hue of the face skin color of Asians and Europeans and Americans is generally between red and yellow. In order to determine the value range of facial skin color that is highly representative in each color component, Douglas Chai selected the face to remove the eyes, lips, and other areas in the experiment, and selected the faces obtained under different lighting, environment, and resolution respectively. The final distribution range of skin color is shown in figure (1). In the CB diagram on the left, it can be seen that the skin color information is concentrated in the abscissa [112133]; In the CR image on the right, the skin color information is concentrated in [140175], so the judgment formula of face skin color can be obtained:

in our study, this threshold is used for skin color segmentation to obtain the binary image of face region

2 skin color region processing

2.1 basic definition of mathematical morphology

the morphological transformation of binary images in mathematical morphology is the processing of sets. The essence of its morphological operator is to express the interaction between the set of objects or forms and structural elements. Therefore, the morphology of structural elements depends on one set, and a batch of experiments can be completed in sequence; Given the shape information of the signal in advance of this operation, morphological image processing is to move a structural element in the image, and then perform intersection, union and other set operations between the structural element and the binary image. Its basic morphological operations are erode operator and flilate. Let B (x) represent structural elements. For each point in workspace e, the definitions of corrosion and expansion are:

corrosion has the effects of reducing the target, increasing the inner hole of the target, and eliminating external isolated noise; Inflation is the process of merging all the background points in the image that are in contact with the target object into the object. The result is to make the target larger and the hole smaller, which can fill the hole in the target and form a connected domain

in mathematical morphology image processing, in addition to the two basic operations of corrosion and expansion, there are two very important operation methods, namely open operation and close operation, which are defined as follows:

2.4 detection of eyes and lips

from the above work, we have determined the position of the face, which reduces the scope of our next detection of human eyes and lips

in this paper, eyes and lips are detected from the perspective of morphology. Let the face area be a, and construct a disc-shaped structural element B with radius R. Close the face area a with the structural element B

the experiment shows that the radius r of the disk structural element is 8 or 9 to achieve the best effect. The effect of the detected eyes and lips is shown in figure (4) and figure (5). From the above figure, we can see that after the structural element B closes the face area a, the face area remains intact, and the human eyes and lips are also saved. Next, we can locate the eyes and lips by finding the center of the white small area

3 analysis and conclusion

the experimental photos of this method are from the photos taken by Internet and digital cameras. Some experimental detection results are shown in figure (6)

this algorithm uses YCbCr color space to segment skin color region. The segmented skin color region is simple and independent, which reduces the interference of background, and uses mathematical morphology to smooth the boundary to facilitate the extraction of face region; In the detection of human eyes and mouth, it is fully conducive to the flexible definition of the size and shape of mathematical morphology template and the characteristics of fast calculation, which speeds up the detection speed, improves the accuracy of detection, and saves time. The results show that the algorithm in this paper can be close to the "waist cut" for accurate recognition and detection of frontal faces, faces with a certain rotation angle and faces in complex backgrounds