Research And Realization Of Lip-reading Algorithm Based On Improved BP Neural Network Algorithm

Lip reading meaning of the speaker’s mouth through to achieve understanding ofthe content of its expression, which can "read" the expression ’s content, read whatwe can do to get the use of visual information lip auditory information throughresearch. Its practical significance is we can help the hearing impaired to get themessage, providing favorable conditions for learning and the exchange and normal.And ways to increase intelligence sources. For our humans, the communicationprocess has always been a process of multi-channel, in accordance with the exchangeprocess can not only talk to each other to determine the content of the other want toexpress ideas, but also with the mouth and body language to better identify each otherwant content expression. Especially in very noisy environments, often rely on eachother’s mouth to identify the contents. From this study lip-reading is very meaningful.Lip-reading is not required by the occurrence as a means of communicationexchanges can be carried out in real-world applications, there are also difficulties. Forexample, due to the different manner of articulation and position as well as theopening and closing levels of the lips, give us lip reading or computer cause somedifficulties. But this does not affect the enthusiasm of our study it.In order to achieve lip-reading function, the computer will go through the lipdetection, feature extraction and recognition multiple steps, we can be divided intothree steps: The first step is to detect lip movement, from the image or video on thelips positioning. The second step is feature extraction, in this step we need effectivefeature extraction lip reading. Third step is lip-reading recognition, feature extractionthrough effective to identify lips.In this paper, taking support vector machine algorithm to identify the lips, thismethod can very well be the face with no other information about the relationship lipselimination, but their accuracy and speed can be guaranteed. Outer lip to lip contourextraction port type is a more feasible approach, and this approach is not affected by head movement. The process of identifying the lips took dynamic clustering algorithmbased on K-means algorithm, and according to this algorithm, lips closed, slightlyopen and gaping validated case work. In order to identify the lip-reading, this articlewas taken neural network algorithm, we propose improvements based on traditionalBP neural network algorithm, making the improved BP neural network algorithm arebetter able to adapt to various conditions, and also improves the efficiency oflearning.