With the development of internationalization,the communication between countries more and more closely, in order to communication easier, there need a voice system which can express all he voice of the phonetic symbol in the world, the international phonetic alphabet arises at the historic moment.however, The study of international phonetic alphabet characters has been in a blank. International phonetic alphabet characters has its own unique characteristics: a various of the types of characters which is not same with others, containing a large number of syllables, and there is a big difference in English characters. In recent years, there are also some other fonts feature extraction method except for the international phonetic alphabet font recognition. recognition process takes longer, the extracted features are complex,and recognition rate is lower, and only for the individual language identification,generality is small.Therefore, research is more and more important for the international phonetic alphabet, especially automatic recognition which is stored in the image fastly,which has a big influence.The study of international phonetic alphabet characters recognition has important academic significance and application value.The recognition of international phonetic alphabet characters including pretreatment,segmentation,feature extraction,and the training of the characters recognition.Feature extraction and recognition of the characters is the research emphasis in this paper.In terms of feature extraction of international phonetic alphabet characters.According to the own unique characteristics of the international phonetic alphabet,extract a variety of features.In this paper, the study of the two characteristics of the international phonetic alphabet characters,which is the structure features and statistical features mainly. Structure features is divided into contour features and moment invariant features. statistical characteristic is divided into based on block projection features as well as the list of local linear characteristics. Experiments show that the mix of features of character recognition has a very high accuracy.In order to improve the effect of character recognition, this paper chose the corresponding pretreatment method, using median filter to remove no use of noise,using the improved Otsu algorithm to binary the image, as well as characters ofskeleton extraction based on mathematical morphology. After verification, the character of the pretreatment method can achieve the desired expectations.Using BP network based on weight for the international phonetic alphabet characters recognition, a single feature recognition effect is not obvious. This paper use a variety of features based on different weights for recognition. The experiment proved that recognition effect is better than single feature.At last,international phonetic alphabet characters recognition system was designed and implemented. |