Stopping Criterion during Rendering of Computer-Generated Images Based on SVD-Entropy

Area of Science:

• Computer Vision
• Image Processing
• Computer Graphics

Background:

• Estimating image quality and noise perception is crucial for image processing and realistic computer graphics.
• Computer-generated images often lack reference images, making traditional noise assessment methods unusable.
• Stochastic noise in global illumination methods requires effective stopping criteria for rendering.

Purpose of the Study:

• To propose a new method for characterizing computational noise in computer-generated images.
• To develop a no-reference image quality assessment (NR-IQA) method.
• To predict visual convergence thresholds for image rendering.

Main Methods:

• Representing image noise using the entropy of Singular Value Decomposition (SVD) of image blocks.
• Utilizing Singular Value Decomposition (SVD)-entropy as input for a recurrent neural network (RNN) model.
• Establishing a relationship between SVD-Entropy and perceptual quality for NR-IQA.

Main Results:

• The proposed method effectively characterizes computational noise using SVD-entropy.
• Recurrent neural networks successfully extract image noise and predict visual convergence thresholds.
• Experimental results show good consistency between the proposed method's stopping criterion measures and psycho-visual scores.

Conclusions:

• The developed NR-IQA method provides a reliable way to assess image quality in computer-generated images without a reference.
• The SVD-entropy and RNN-based approach offers a promising solution for determining optimal rendering stopping criteria.
• This research contributes to improving the efficiency and quality of photo-realistic image generation.