Research And Realization Of The Face Detection System Based On DSP

Face detection is one of the important researching aspects of human-machine interaction based on the image processing, which means that find and locate all the human faces in the still images or video sequences in spite of the variation of the faces' location, size, orientation, posture and the illumination in the images. Besides, as a key technique in the human face information processing, face detection has the increasingly extensive application of identity recognition, content based retrieval, automatic monitoring, human-machine interaction and so on. At present it has become a research hotspot in the field of pattern recognition and computer vision.The paper marks out the system's general frame of hardware and software in the researching procedures of this system. Concerning the hardware, this system is not based on a normal PC used for sheer arithmetic research, but on the system platform formed by the DSPs(Digital Signal Processors)- choose the chip TMS320C6416 produced by the TI Corporation as the processor- to construct an applied hardware system. On the other hand, the software integrate development environment of this system is CCS(Code Composer Studio) 6000 2.0 which contains the real-time operation system(RTOS) which could be applied to build tasks and interruptions in the program conveniently and manoeuvre the tasks through the signal and mailbox in order to meet the requirements during the exploitation. During programming, the code was designed in accordance with the C64XX software development flow. After this, the debug tools provided by the CCS was imposed to analyze the performance of C code, and then according to the hardware property of the DSPs chips we take relevant method in this paper to optimize the program. The optimization improves the parallel utility grade of hardware resource, and consequently elevates the codes' running speed to meet the real-time requirement.In the aspect of the arithmetic, this work expatiates the essential principle of the system and the current face detecting methods according to the system's hardware, and then choose the skin-color detecting method as the basic means. In the moment of image pretreatment, the light compensation, namely histogram equalization, is used to compensate the chromaticity component of video frames, and this method greatly increases the accuracy rate of the later procedure of detection. This work also builds the human face skin-color model, and uses it to binary images of the video sequence in the detecting moment. Subsequently, the morphological filtering method is used to deal with the images to get the ones with higher quality. On the basis of the former steps, the human face area would be affirmed via the marking connected-components algorithm.The experimental results indicate that the system can detect human face by rule and line correspondingly, and meet the real-time demand. If the face detection technique and traditional video monitoring systems were combined, the promotion of the level of intelligence of traditional systems could be made, and the base of the consequent development of human-machine interaction technique would be established.