English Word Recognition System Based on BP Neural Network

In order to solve the problem that English word recognition rate is not high in the process of globalization, an English word recognition system based on BP neural network is proposed. The system studied the character preprocessing, feature extraction, and classification recognition algorithms in the image. In the character preprocessing, the original image was successively compressed, gray processing, binarization processing, orientation correction, character segmentation and normalization processing. Bilinear interpolation algorithm is used to compress the original image. The gray weight of OpenCV open library is used for grayscale. In the binarization processing, the traditional methods are compared and studied. Different binarization processing methods are designed for different images, and the traditional Bernsen algorithm is improved. The original image and the Gaussian filtered image are calculated by Bernsen algorithm. The results are as follows: for bad images such as uneven lighting and bad background, the recognition rate of the improved Bernsen algorithm is about 81%, the recognition rate of the OTSU algorithm is only 53%, and the standard Bernsen algorithm is only 75%.The recognition rate is around 49% for a single mesh, 54% for a single penetrating feature, and 87% for a binary combination. The recognition rate of double feature is higher than that of single feature, which is 38% higher than that of single coarse mesh feature and 32% higher than that of single penetrating feature. In the actual experiment, the average recognition rate of printed letters is 86.2%, while the recognition rate of handwritten letters is 73.2%. The application of this system through OCR technology to realize the conversion from pictures to text, can directly identify English words from pictures and read dictation, easy to use, can achieve the purpose of auxiliary teaching, help dictation, has important practical significance.