Abstract With the joining of digital media and Internet, the application of the conversion of cross-media information becomes wider and wider, which makes the integration of multimedia information possible. The transformation of image information to sound information is also in the stage of exploration. It can be applied to all aspects of the industry. Most importantly, it can be used to guide the blind. In this paper, an improved image-to-sound algorithm based on original method is proposed to judge whether there are obstacles in front of the road and test the sound segment through human ears. Among different mapping methods, the highest accuracy can reach 84.2%, which is much higher than the original method. At the same time, an objective evaluation algorithm is designed based on the neural network, which can replace the human ear to judge a large number of image information, and the matching degree of this model can reach 68%.