Jellyfish Chicken Swarm Optimizer based Convolutional Neural Network with transfer learning for Recognizing Text in Complex Images

Abstract

Text acquired in images contains diverse information and thus it is widely utilized in several applications for understanding image scenarios. Also, it is employed for retrieving visual information. Semantic information in complex image is highly important for the humans to recognize the complete environment. Even though, text present in images reveal the flexible form in unconstrained state that makes identification of text as well as character recognition process as very challenging task. Here, Jellyfish Chicken Swarm Optimizer based Convolutional Neural Network with transfer learning (JCSO_CNN with TL) is presented for text recognition in complex images. The input image obtained from database performs text foreground extraction employing GrabCut method. Thereafter, text detection is carried out by Differentiable Binarization Net++ (DBNet++). Afterwards, character segmentation is accomplished utilizingFuzzy Local Information C-Means (FLICM). Finally, character recognition is conducted using CNN with TL in which CNN is employed with hyperparameters from the trained models namely LeNet-5. Moreover, CNN with TL is tuned by designed JCSO that is formed by incorporating Jellyfish Search Optimizer (JSO) with Chicken Swarm Optimization (CSO). Furthermore, JCSO_CNN with TL attained 90.8% of precision, 94.7% of recall and 92.7% of f-measure.