A Study On Image Characteristics Recognition Method And Its Application In Deaf Visual Recognition

With the fast development of computer and information technology, 3G(the 3rd-Generation) life has gone homes, and 4G(the 4th-Generation) is coming to the forefront. These high and new technologies have brought earthshaking change to people and it made exchange and communication between people change more speedy and convenient. When we enjoy these advantages brought by information, we should pay more attention to disadvantageous group as important constituent part of our society, namely, the deaf people. Now there have more than 2,000 thousands deaf-and-dumb peoples in China. In order to make them infuse normal social life better, and eliminate communication obstacle. The more and more national experts and scholars focus on the deaf visual recognition research.Now, the existing most methods of deaf visual recognition focus on the sign language recognition algorithm based on data gloves, and the visual recognition algorithm based on image process. A lot of researches are to combine with the two algorithms. It firstly collected pictures by the former, then processed and recognized pictures by the latter so as to obtain better recognition effect.Although we have made some progresses in the research field of deaf visual recognition, we still face many challenge tasks, such as invariable characteristic extract of hand shape, transition model research of hand gesture and so on. These problems resulted in many shortages in sign language recognition rate and algorithm robustness. Aiming at the mentioned above, the paper has made some researches and obtained a phase research achievement. It has been applied into the application software development of the developed deaf people communicate visual recognition system undertaken by author, and the part of the research results has been put into the actual teaching trial. Although the application software obtains better application effect, it still needs to make improvement. The main research works and innovations of the paper are concluded as the following:â‘  Aiming at the shortages of watershed algorithm in noise sensitive which results in edge discontinuous, forge edge and over segmentation phenomenon, the paper proposed a kind of recognition method of image geometry feature which based on morphology and watershed algorithm. The method is based on gray Level feature of the image, according to segment at maximum degree of foreground and background, it can automatically ascertain segmentation optimal threshold of the image by forthright means. Then it further restricted path cost function of traditional watershed algorithm according to the segmentation threshold. The essence of the method is to narrow search scope, so increase implements speed of the algorithm.â‘¡ Aiming at some puzzles of existing various fusion algorithms in image recognition, such as light spectrum twist, algorithm complexity, needed more memory, and difficultly taken into account both total structure and details and so on, the paper proposed an image fusion method which is based on bi-orthogonal wavelet transform and texture consistency measure. The method is respectively to make wavelet decomposition for source image by means of bi-orthogonal wavelet transform; choose wavelet coefficient of low frequency according to a certain proportion to compose wavelet coefficients of low frequency of fusion image. For the high frequency coefficient, the method is to analyze the specific areas edge characteristic of various coefficients of high frequency by means of texture homogeneity measure, so as to determine high frequency coefficients matrix of fusion image according certain rules. Then the method makes some process to fuse the images. Experimental results have shown that the method not only can distinguish the forge edge better, make detail information change richer and truer, but also can balance total visual effect. It could get better recognition effect.â‘¢ Aiming at the shortage of recognition of hand shape, the paper proposed sign language recognition algorithm based on fuzzy BP neural networks. Firstly, the method makes fusion calculation on hand shape image and lip shape image collected by means of fuzzy BP neural networks. The fusion results respectively view as the fuzzy sets of hand shape and lip shape. Then it makes calculation on the fuzzy operators, matches the calculation results and database of sign language symbol, and makes fuzzy operation on the gotten two sign language sets further, so as to get the finally recognition result. Experimental results have shown that the method is effective.â‘£ Aiming at the shortage that in the complex illumination, traditional human face characteristic recognition algorithm often ignored contrast of local area, discard part important texture information, so that the recognition effect is undesirable, the paper proposed an improved LBP human recognition method. Firstly, the method makes illumination normalization processing, control illumination change to certain extent. Then, it maps contrast value of local pixel to an interval range by means of improved LBP, makes the contrast value of local pixel resulted from illumination change to a controlled interval, so that the method could make the image possess illumination invariable and it could recognize human face characteristic better.