Tile-boundary artifact reduction using odd tile size and the low-pass first convention

Area of Science:

• Image processing
• Digital signal processing
• Computer vision

Background:

• Wavelet-based lossy image coding commonly exhibits tile-boundary artifacts.
• The underlying causes of these artifacts have remained poorly understood.
• Conventional methods for tile size selection and symmetric extension contribute to artifact formation.

Purpose of the Study:

• To identify the root causes of tile-boundary artifacts in wavelet image decomposition.
• To propose a novel method for mitigating these artifacts.
• To demonstrate the effectiveness of the proposed method in reducing visual distortions.

Main Methods:

• Analysis of conventional tile size selection and symmetric extension techniques.
• Development of a novel image decomposition method utilizing odd tile sizes (2N + 1 samples).
• Implementation of the odd tile length low-pass first (OTLPF) convention.

Main Results:

• Boundary artifacts are an inherent consequence of standard tile size selection and symmetric extension.
• The OTLPF convention with odd tile sizes demonstrably reduces boundary artifacts compared to even tile sizes.
• Images compressed using the OTLPF method exhibit significantly fewer artifacts at equivalent bit rates.

Conclusions:

• The choice of tile size and symmetric extension method critically impacts artifact generation in wavelet image coding.
• Employing odd tile sizes, specifically with the OTLPF convention, offers a viable solution for artifact reduction.
• The OTLPF convention is compatible with existing standards like JPEG 2000 Part 2, enabling practical implementation.