Research On Key Technologies In Blind Guidance System With Video Input And Auditory Display

People observe surroundings and acquire knowledge through vision since they were born. Ninety percent of information is got through vision. So visual damage is regarded as the greatest misfortune to the human life. Researchers are focusing on how to get visual information through other sense organs. The key to this problem is to translate visual information to the signals which can be sensed by other sense organs. The equipment with this function is called blind guidance system.With the progress of computer vision and digital image processing, the blind guidance system with video input and auditory display (VIAD) is widely studied because it inputs more information and it is noninvasive. But there are two problems in traditional VIAD blind guidance system as follows. First, the algorithms of detecting obstacles are the main algorithms. The information except obstacle information and reasoning ability of the blind are ignored. Second, primary features of the image are chosen and mapped in mostly algorithms. The sound produced in this mapping mode is complicated and difficult to understand and use for the blind.Aiming at these problems, this paper presents useful methods of image processing and auditory display based on the mapping of static image features in order to make the blind use the VIAD blind guidance system easily and get more plentiful information. The main works are as follows:(1) Aiming at the amount of imbalance between objects and backgrounds, a support vector machine model with dynamic penalty factors (DC-SVM) is presented. The penalty factors are directly related to the ratio of each class of samples to total samples. This model can effectively improve the correct classification rate for the class of few samples. Then DC-SVM is used to segment objects in the images. After the backgrounds are removed, the objects are mapped into the sound with multiple frequencies. And the blind can acquire environmental information via the mapped sound after a period of learning and training.(2) In order to solve the problem that details become blurred when the images are enhanced by the anisotropic diffusion with shock filter (ADSF), a gradient controlled anisotropic diffusion with shock filter (GCADSF) is presented. This model uses an exponential function which monotonically decreases with the gradient modulo to control the diffusion in the vertical direction of the image gradient. So it protects the image details as the edges are enhanced and the noise is filtered. The image is enhanced in the reduced dimension by lifting wavelet transform combined with GCADSF to meet the needs of the multi-scale music display algorithm. And then this image is mapped into musical notes through the multi-scale music display algorithm. The sound is pleasing to ear and enjoyable.(3) According to the characteristics of receptive field, an image edge detection method is presented based on the reconstruction of edge basis functions. Firstly, the image is decomposed into the linear combination of ICA basis functions. Secondly, ICA edge basis functions are obtained by a typical edge image. Finally, the edge image is got via the reconstruction of edge basis functions. This edge detection method accords with vision characteristics because ICA basis functions simulate the characteristics of receptive field very well. And an edge detection method is presented based on wavelet transform and Canny algorithm. In this method, Canny algorithm is used to detect the edges in the low-frequency image of wavelet transform. This method can reduce the image dimension and effectively restrain the effect of image noises. Binary edge image is mapped into MIDI music with a variety of timbre and pitch via MIDI music display algorithm. This blind guidance algorithm achieves high recognition rate when it is used to help the blind to recognize geometrical figures in the image.(4) Itti visual attention model is applied to the research of blind guidance algorithm. Regions of interest (ROI) are considered to include important image information. So the color, position and size are extracted and mapped. According to the color distribution of ROI, a method called filter, dilation and histogram algorithm (FDH) is introduced in order to extract the main color of ROI. The position of sound is used as the mapping parameter. The color, position and size of ROI are mapped into a musical note which comes out from the specific direction and have a specific pitch. This blind guidance method chooses the most important information in the image to map for the blind. And it is an effective algorithm.