Improved Alexnet-Based Character Detection Method

Extracting text information in complex images is a hot spot in pattern recognition research with broad application prospects. Natural scene door numbers produce serious distortion due to blurred images, uneven illumination and low light illumination, which makes it difficult to achieve ideal results in character recognition, and recognizing characters of arbitrary length is even more of a challenge. In this paper, we adopt the improved Niblack's local threshold segmentation method to segment the images, and mark the connected areas of the segmented images to highlight the important features, and finally input the above pre-processed images to the improved AlexNet network for target detection on SVHN (Street View House Number) to realize the detection of characters in real scenes. The experimental results show that the improved Alexet-based target detection method is able to complete the detection task of streetscape door number characters well with 92.89% correct recognition rate.