Scalable Image Coding Based on Epitomes.

Area of Science:

• Computer Vision
• Image Processing
• Video Coding

Background:

• Scalable image coding is crucial for efficient video transmission across diverse networks.
• Existing methods like the Scalable extension of HEVC (SHVC) face challenges in optimizing rate-distortion performance.
• Epitomes offer a novel factorized image representation with potential for compression gains.

Purpose of the Study:

• To propose a novel scalable image coding scheme utilizing the concept of image epitomes.
• To focus on spatial scalability by transmitting only the epitome in the enhancement layer.
• To restore missing pixels in the enhancement layer using learning-based super-resolution techniques.

Main Methods:

• An epitome-based scalable image coding scheme is developed.
• The enhancement layer consists solely of the image epitome.
• Two local learning-based super-resolution methods are employed for pixel restoration: locally linear embedding and linear mapping between base and epitome patches.

Main Results:

• The proposed scheme achieves significant improvements in rate-distortion performance.
• Experimental results demonstrate superior performance compared to the Scalable extension of HEVC (SHVC).
• The method effectively reconstructs enhancement layer pixels from the epitome and base layer information.

Conclusions:

• The epitome-based approach offers a promising direction for advanced scalable image coding.
• The integration of super-resolution techniques enhances the efficiency of the enhancement layer.
• This novel scheme provides a competitive alternative to current standards for scalable video compression.