Prediction of image partitions using Fourier descriptors: application to segmentation-based coding schemes

Area of Science:

• Computer Vision
• Image Processing
• Video Coding

Background:

• Partition sequences are crucial in segmentation-based video coding.
• Accurate prediction of these partitions is essential for efficient compression.
• Existing methods may lack robustness in handling complex motion and shape deformations.

Purpose of the Study:

• To develop a novel prediction technique for partition sequences.
• To apply this technique for partition interpolation and extrapolation in video coding.
• To improve the efficiency and accuracy of segmentation-based video compression.

Main Methods:

• A region-by-region approach involving parameterization, prediction, ordering, and creation.
• Utilizing Fourier descriptors to model regular motion and shape deformation over time.
• Predicting region evolution in the Fourier domain and combining them using morphological tools.

Main Results:

• Demonstrated effectiveness in noncausal partition prediction for interpolation tasks.
• Showcased successful causal partition prediction for extrapolation in coding applications.
• Validated performance on complete partitions and sequences of binary images (Video Object Planes - VOPs).

Conclusions:

• The proposed technique offers a robust method for predicting partition sequences.
• It effectively handles both regular motion and shape deformation for improved video coding.
• The approach shows promise for enhancing segmentation-based video compression strategies.