No-Reference Depth Assessment Based on Edge Misalignment Errors for T+D Images.

Area of Science:

• Computer Vision
• Image Processing
• 3D Reconstruction

Background:

• Accurate depth data is essential for many applications, but obtaining perfect ground truth is challenging.
• No-reference quality assessment for depth maps is highly desirable due to the difficulty of acquiring error-free ground truth.

Purpose of the Study:

• To develop a novel no-reference depth quality assessment scheme.
• To focus on detecting and quantifying depth edge misalignment errors in texture-plus-depth (T + D) images.
• To create a robust metric for evaluating depth map quality based on detected misalignments.

Main Methods:

• A novel scheme focusing on depth edge misalignment errors in texture-plus-depth (T + D) images.
• Detection of misalignments by matching texture and depth edges using spatial similarity, edge orientation similarity, and segment length similarity.
• Edge matching performed on segments for robust detection, leading to a no-reference quality metric.

Main Results:

• The proposed scheme accurately detects depth edge misalignment errors.
• The developed no-reference depth quality metric shows high consistency with full-reference metrics.
• The metric correlates well with the quality of synthesized virtual views.

Conclusions:

• The novel scheme provides an effective no-reference method for depth quality assessment.
• The detected misalignments can be utilized to enhance depth quality in T + D applications.
• This approach addresses the need for reliable depth quality evaluation when ground truth is unavailable.