Sharmila Chidaravalli
Department of Information Science & Engineering, Global Academy of Technology, VTU, Bangalore, Karnataka, India.

Vimuktha E Salis
Department of Information Science & Engineering, Global Academy of Technology, VTU, Bangalore, Karnataka, India.

DOI https://doi.org/10.33889/IJMEMS.2026.11.5.088

  ;

Abstract

With the surge of process modernization and enhanced data management, the world is advancing towards digitization and automation more than was presumed. The Internet is becoming one of the most favored means of communication and multimedia, coupled with data compression methods, plays a crucial role in controlling the huge volume of data that is being produced on a regular basis. Modeling novel methods for the automatic analysis of compressed data without decompression is pressing requirement of the hour. In this work a deep learning method called Linear Activated Cosine Deep Neural Network based text feature extraction from Fuzzy Filter (LACDNN-FF) quality improved compressed image documents is proposed. The LACDNN-FF method is modeled using an agent AI model. To be more specific, three different agent AIs, namely the Image Enhancement Agent (IEA), Layout Analysis Agent (LAA) and Text Feature Extraction Agent (TFEA) are employed in the design of the LACDNN-FF method. Initially, with the scanned image dataset of compressed image documents as input, two dimensional flexible fuzzy filter-based pre-processing is applied by the IEA with the intent of eliminating noise, thereby generating quality-enhanced compressed documents. Next, Linear Activated Discriminant Analysis with Deep Neural Network-based Segmentation is applied by the LAA to perform document segmentation and classify the contents using pixel-precise semantic segmentation. Finally, Cosine CNN-BiLSTM (Convolutional Neural Network - Bidirectional Long Short-Term Memory) neural network based Text Feature Extraction is applied by TFEA for accurate text extraction and text feature processing. The proposed deep learning method LACDNN-FF outperforms existing methods and shows a promising solution for accurate text feature extraction with minimal error rate.

Keywords- Two dimension flexible fuzzy filter, Linear activated discriminant analysis, Deep neural network, Cosine convolutional neural network, Bidirectional long short-term memory.

Citation

Chidaravalli, S., & E Salis, V. (2026). Deep Learning Based Cosine Text Feature Extraction from Fuzzy Filter Quality Improved Compressed Image Documents. International Journal of Mathematical, Engineering and Management Sciences, 11(5), 2198-2227. https://doi.org/10.33889/IJMEMS.2026.11.5.088.