An Optimized Ant Colony Algorithm for Text Edge Extraction

The deep learning-based detection model has achieved outstanding results in the field of natural scene text detection. However, the lack of image detail information in the depth feature will result in the inseparable text objects in the detection results cannot be separated, and will affect subsequent text recognition. Precision. In order to solve this problem, this paper proposes an optimized ant colony algorithm for extracting text edges, and combines the optimized ant colony algorithm with the depth detection model to achieve the purpose of optimizing text detection and recognition accuracy. The algorithm uses the gradient information of image partition to initialize the position of ant colony, and optimizes the process of ant colony searching for edges according to the characteristics of text objects. An adaptive pheromone dissipation coefficient is used to deal with the ant colony movement process, which effectively solves the traditional ant. Group algorithms are slow to converge and sensitive to noise. This paper will conduct experiments on the standard text detection sets of ICDAR 2011 and ICDAR 2013. The results show that combining the optimized ant colony algorithm with the FCN-based deep text detection model can not only effectively improve the accuracy of the original deep text detection model. The text recognition accuracy can be further optimized.