Deep Concatenated Residual Networks for Improving Quality of Halftoning-Based BTC Decoded Image

Area of Science:

• Image Processing
• Computer Vision
• Deep Learning

Background:

• Halftoning-based block truncation coding (H-BTC) offers improved decoded image quality over classical block truncation coding (BTC).
• H-BTC decoded images frequently exhibit impulsive noise, degrading visual perception.
• Addressing this noise is crucial for practical applications of H-BTC.

Purpose of the Study:

• To propose a novel deep learning-based technique for suppressing impulsive noise in H-BTC decoded images.
• To enhance the overall quality of H-BTC decoded images.
• To provide an effective solution for an ill-posed inverse imaging problem in image compression.

Main Methods:

• Utilizing convolutional neural networks (CNNs) for noise suppression.
• Employing residual learning frameworks to improve image reconstruction.
• Applying a deep learning approach to address the ill-posed nature of noise removal.

Main Results:

• Significant reduction in impulsive noise occurrence in H-BTC decoded images.
• Demonstrated improvement in the subjective and objective quality of the processed images.
• Validation of the proposed method's effectiveness through experimental evaluations.

Conclusions:

• The proposed deep learning method effectively suppresses impulsive noise in H-BTC decoded images.
• The technique enhances image quality, leading to improved visual observation.
• CNNs and residual learning provide a robust solution for improving H-BTC image compression quality.