Deep features based convolutional neural network model for text and non-text region segmentation from document images

A deep convolutional neural network model is presented here which uses learning features for text and non-text region segmentation from document images. The key objective to extract regions the complex layout images without any prior knowledge of segmentation. In a real-world scenario, or magazine contain various information along with such as symbols, logos, pictures, graphics. Extraction challenging. To mitigate these issues, an efficient robust technique has been proposed in this paper. implementation divided into three phases: (a) method pre-processing using different patch sizes employed handle situations variants fonts mage; (b) predict ambiguous within image; (c) post-processing image situation where by utilizing recursive partitioning those their proper classes (i.e. non-text) then system accumulates responses predictive patches varying resolutions handling variations image. Extensive computer simulations have conducted collection Google sites ICDAR 2015 database. Results are collected compared state-of-the-art methods. It reveals that more effective • analyze architecture proposed. deals case demonstrate performance. findings comprehensive manner.