Recognition of Video Time Watermark Based on Water-Wave Connected Domain Segmentation Algorithm
Pengfei Meng, Shuang-cheng Jia, Qian Li · 2021 IEEE 21st International Conference on Communication Technology (ICCT) · 2021
Character segmentation has always been a difficult task in the field of text recognition. The common connected domain segmentation algorithm not only repeatedly traverses the pixels, but also does not solve the problem of character adhesion segmentation. In this work, targeted at these problems, we proposed a time watermark recognition method, which can quickly realize the time watermark recognition only by labeling a few sample pictures. It is very suitable for the scene where time watermark needs to be recognized quickly. Compared with the traditional connected domain segmentation algorithm, the water wave connected domain segmentation algorithm segmented characters more accurately, and it can also handle the situation of light adhesion. The segmentation method proposed in this paper avoids the repeated stacking and popping or recursive operations of the traditional connected domain segmentation algorithm mostly, and provides a new idea for character segmentation. Based on the water wave connected domain segmentation algorithm proposed in this article, four different image processing algorithms are compared and studied: template matching [1], Hamming distance calculation [2], Euclidean distance calculation [3], and pixel comparison calculation. The recognition method proposed in this paper is compared with the CRNN [4] recognition method under the same conditions, and the results show that when a small number of samples are used, the recognition rate of the proposed method reaches 92.3%, and the CRNN recognition rate is 0. Compared with CRNN, the recognition method proposed in this article has achieved good results on very few samples.