Noise reduction of VQ encoded images through anti-gray coding

Area of Science:

• Digital Image Processing
• Information Theory
• Computer Vision

Background:

• Vector Quantization (VQ) is a widely used image compression technique.
• VQ encoded images are susceptible to channel errors, leading to noise and reduced image quality.
• Existing noise reduction methods may not be optimal for VQ encoded data.

Purpose of the Study:

• To develop an effective noise reduction scheme for VQ encoded images.
• To propose a novel anti-gray coding (AGC) technique for improved error resilience.
• To introduce a robust channel error detection and correction mechanism tailored for VQ.

Main Methods:

• Implementation of anti-gray coding (AGC) by assigning binary indices to codevectors to maximize distance between neighbors.
• Image classification into uniform and edge regions for targeted noise detection.
• Development of a mask for error detection based on image classification and AGC characteristics.
• Mathematical derivation of an error detection criterion.
• Error correction by selecting recovered indices that minimize gray-level transitions.

Main Results:

• Achieved a probability of error detection greater than 86.3% at a random bit error rate of 0.1%.
• The probability of undetected errors is less than 0.1%, with invisible artifacts.
• Image quality improvement of 3.9 dB compared to images encoded solely with AGC.

Conclusions:

• The proposed AGC and noise detection/correction scheme effectively reduces noise in VQ encoded images.
• The technique offers a high probability of error detection and correction with minimal visual impact.
• This method significantly enhances the robustness and quality of VQ encoded images in the presence of channel noise.