Image Processing Apparatus Using Intermediate Image Artifact Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods face challenges in accurately and efficiently deconvolving blurred images due to noise and variability in blur kernels, limiting their accuracy and calculation speed.
Innovation Solution
A machine learning method using a convolutional neural network that generates an intermediate image with artifacts less sensitive to blur kernel changes, allowing for artifact removal and deconvolution without direct learning of the deconvolution process, and improving the accuracy of image correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning techniques are used for deconvolution, then artifact removal capability is improved, but calculation speed deteriorates
Solution Approach 1:
The patent segments the deconvolution process into two distinct stages: (1) generating an intermediate image with artifacts that are less sensitive to blur kernel changes, and (2) removing artifacts using a pre-trained artifact removal model. This segmentation allows the system to avoid real-time deconvolution calculations while maintaining high artifact removal accuracy, thus resolving the contradiction between precision and productivity.
Solution Approach 2:
The patent performs preliminary action by pre-training the artifact removal model using paired datasets of intermediate images and ground truth images before actual image processing. This pre-training enables the model to quickly remove artifacts from new images without requiring computationally intensive real-time deconvolution, thereby improving calculation speed while maintaining accuracy.
2Adaptability or versatility
If deconvolution method changes according to blur kernel, then adaptability to different blur types is improved, but calculation complexity deteriorates
Solution Approach 1:
The patent creates a universal artifact removal model that can handle multiple types of blur kernels (Gaussian blur, motion blur, defocus blur) without requiring separate models for each blur type. The model is trained on diverse datasets containing various blur kernels, enabling it to adapt to different blur conditions while maintaining a single, unified processing pipeline that reduces calculation complexity.
Solution Approach 2:
The patent introduces an intermediate image as a mediator between the blurred input image and the final restored image. This intermediate image contains artifacts that are less sensitive to variations in blur kernels, allowing a single artifact removal model to effectively handle different blur types without requiring complex adaptive deconvolution algorithms for each blur kernel type.
3Adaptability or versatility
If noise is present in blurred images, then real-world applicability is improved, but deconvolution difficulty deteriorates
Solution Approach 1:
The patent performs preliminary action by pre-training the artifact removal model on datasets that include noisy blurred images with various noise levels and types. This pre-exposure to noise during training enables the model to learn robust artifact patterns that persist despite noise, allowing it to effectively remove artifacts from real-world noisy images without requiring complex noise filtering or adaptive deconvolution techniques.
Data Source
AI summary
Disclosed are an image processing apparatus and an image processing method. The image processing method performed in the image processing apparatus comprises receiving a blurred image; generating an intermediate image from the blurred image, the intermediate image including artifacts with less sensitivity to changes in a blur kernel than the blurred image; and generating a first corrected image by removing the artifacts of the intermediate image using an artifact removal model. Therefore, it is made possible to learn how to remove artifacts of the intermediate image without the image processing apparatus learning a direct deconvolution method.