L0 Image Smoothing With Deep Gradient Prior for Edge Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods for computer vision face challenges in efficiently removing unnecessary information and preserving high-frequency components, leading to computational complexity and performance degradation in edge preservation.
Innovation Solution
A one-step gradient minimization method using a deep neural network (GNet) to improve image sharpness by minimizing gradients of pixels, excluding critical edge areas, and performing smoothing through one-step estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If global-based edge preservation methods (L0, L1-norm, weighted least square) are used to preserve edges while minimizing objective functions, then edge preservation performance is improved, but processing speed deteriorates due to increased computation as image size increases
Solution Approach 1:
The image is divided into multiple blocks, and the 5-direction gradient computation is performed in parallel for each block using GPU architecture. This segmentation enables the global-based L0-norm smoothing to process large images efficiently by distributing computation across multiple independent blocks simultaneously.
Solution Approach 2:
The patent replaces traditional CPU-based sequential computation with GPU-based parallel computation architecture. By utilizing the massive parallel processing capability of GPUs, the computationally intensive L0-norm smoothing with 5-direction gradients achieves significant speedup while maintaining edge preservation quality.
2Manufacturing precision
If Cho's improved L0-norm smoothing method is used that adds directional differential operator and processes 5-direction gradient in parallel, then smoothing performance is improved, but computational complexity increases
Solution Approach 1:
The computation domain is segmented into multiple non-overlapping blocks, allowing independent parallel processing of 5-direction gradients in each block. This reduces the memory access complexity and enables efficient GPU utilization without requiring complex global synchronization.
Solution Approach 2:
The patent combines multiple gradient directions (5-direction: horizontal, vertical, and 4 diagonal directions) into a unified L0-norm optimization framework. By merging these gradient computations into a single parallel processing pipeline on GPU, the method achieves improved smoothing performance without proportionally increasing computational complexity.
3Manufacturing precision
If structure-guided 0-optimization method with two 0-norm regularization terms is used to preserve unnecessary edges, then edge preservation performance is improved, but processing speed deteriorates due to increased amount of computation
Solution Approach 1:
The patent extracts and processes only the essential gradient information (5-direction gradients) within each block independently, removing unnecessary global computations. This extraction of key features enables efficient parallel processing while maintaining effective edge preservation through the L0-norm regularization.
Data Source
AI summary
In an embodiment of the present inventive concept, there is provided an 0 smoothing method performed on the basis of deep gradient prior information to improve sharpness of an image by an image quality improving device, and the method comprises: a gradient-improved image generation step of generating a gradient-improved image by minimizing the gradients of pixels of an original image, by the image quality improving device; and a smoothing-improved image generation step of generating a smoothing-improved image smoothing-processed through one-step (0) estimation on the gradient-improved image, by the image quality improving device.