Research And Implementation Of Face Detect In Compressed Domain

With the continuous development of computer technology , multimedia technology and network technology has its unique situation to become the dominant era. Internet enriches the daily lives of human beings, to provide mankind with the leisure and entertainment "venue". Moreover, in large part to meet human the desire to world civilization and rich information resources . Subsequently developed e-commerce, online banking, personal identity authentication, information security, automated identity verification has received increasing attention, and are slowly deep into more areas of society, so the security requirements needed increasingly improved. Traditional authentication (such as magnetic card, certificates, etc.) is increasingly demonstrating its disadvantages, such as easily lost, or pretending to be forged, etc. So that the staff does not recognize a person's true identity, thus giving it the appropriate permissions, allowing it to complete a certain operation. With the depth study of field of computer vision field, scientists found that people face as the objects for identification and detection has a very bright future. Face is a human biological characteristics, with strong self-stability and individual differences. Therefore, it would be the ideal basis and means serving as the authentication. However, the development of multimedia technology to promote exponential growth of multimedia data, making image storage and transmission have become somewhat reactive. To face the intensifying the image information processing age, scientists have input a lot of energy to study it, and put forward face detection technology based on compressed domain.In the information age of explosive growth, large amounts of data, video, audio, images and other information influx of people's lives. The traditional mode of information processing in the pixel domain is coped by its digital processing computer, with a large amount of data processing mode, and there is a compression format on a wide range of information is difficult to achieve efficient processing. Therefore, image storage and transmission process, if there is a compressed format that will bring convenience to the whole process。At present, the digital image compression techniques can be divided into two main lines: Based on the discrete cosine transform (DCT transform) image compression technology, and wavelet-based image compression technology. Based on DCT (Discrete Cosine Transform) transform image compression is now widely used in our reality image compression technology, because of its good energy compaction properties and fast computing power, has been widely used in image compression field. Also, because DCT-based analysis, handling features such as simple and so on, International Static and dynamic image compression standard JPEG image compression standard MPEG both use DCT transform, therefore, the DCT-based image coding compression technique is a very popular subject, but also accelerated the development of this area. Image Compression Based on Wavelet Transform can be used as a high-bit-rate image compression technology to solve some graphic information has brought a large number of digital image data, speed up communication transmission speed and ease shortage of network bandwidth. For the wavelet transform can be said that these advantages of wavelet transform in this area has a very good application prospects. Therefore, the wavelet transform in our country is bound to have a good role in promotion and application.This thesis introduces the knowledge including the face detection technology and image compression technology, and making research focus on the compressed domain face detection technology and put forward an improved wavelet-based DCT compressed domain face detection technology.Nature of multimedia data are jumbled, disordered, unstructured characteristics, while in the calculation of resource-limited environment, in the multimedia data structure and content of the information is not be lost under the premise of its storage and transmission is extremely difficult to . Traditional approach is based on the spatial domain is still very common, and that it is first compressed image decoding converted to the spatial domain, and then each pixel processing, and finally deal with the image re-encoding into a compressed image. Operation of this approach is too big, and additional links to increase understanding of compression. To solve this problem, we have proposed compressed domain based image compression, image compression domain without decompression or partial decompression of the circumstances under which processing, processing speed, efficiency and ensure the security of multimedia information.Face research in computer vision and pattern recognition has two main factors:face recognition and face detection. At first, face mainly research in face recognition, and recognition algorithm is the traditional assumption has been made in a positive face, or under the conditions that face is available. However, this assumption of the expanding range of applications and increasing demand for the development of the actual system conditions, existing many disadvantages at the same time. Thus, the face detection have been studyed and developed as a separate content by people . Now, there are many people face detection algorithms have been incomed to which domestic and foreign literature.This paper presented the background on the subject were briefly introduced, and face detection technology's research and development made simple to understand whether domestic or foreign. On the basis of the existing theories do further research and discussion. In the DCT domain, the neural network, wavelet transform theoretical knowledge of image processing were used in order to better face on the testing samples.Then selected low frequency coefficients as the feature vectors from the DCT transform coefficientd. In the space of eigenvector, the employer faces the model cluster clustering and non-face samples to be addisted to the recognition of human face model. Final results were verified by experiments. Even many of the major international conferences and journals on the issue of face detection is also very concerned about.Detected to the human face in the compressed domain , some of the characteristics are vulnerable to external factors such as hair, side faces and so on. And simple to use BP training samples , the system's convergence speed is slow, the system prone to instability phenomenon, sometimes it will fall into the error function of the local minimum. In order to solve these problems, this thesis adopts the wavelet transform and genetic algorithm optimization neural network approach. First face image is decomposed by wavelet transform can obtain the original image by low-frequency sub-image, which describes facial expressions and gestures of the same features, with better stability, but also wavelet transform has effectively reduced the wavelet transform image the number of dimensions. For the neural networks, we are using the genetic algorithm optimization was carried to them. Genetic algorithm has good robustness in the global solution space, from multiple regions to find the optimal solution to avoid the lingering problem of local optimum. Experimental results show that: with only use network classifier compared , use of wavelet transform, exploit its excellent time-frequency domain partial performance, the recognition rate of the system significantly improved, and the calculation has great advantages. To avoid some missed and mistakes prosecution. Compared with the original algorithm, it has a higher correct detection rate and a smaller time complexity.