Aberrant Pixel Reconstruction Through Focused Neural Network Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies fail to effectively correct aberrant pixel values caused by dysfunctional pixels in image sensors, leading to visible defects in reconstructed images due to the rarity and impact of these malfunctions being overlooked by conventional cost functions.
Innovation Solution
A neural network training method that generates modified images with aberrant pixels, applies a convolutional neural network for attention data generation, and uses a focused cost function with backpropagation to correct pixel values, enhancing image reconstruction by identifying and addressing dysfunctional pixels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional cost functions are used in image reconstruction, then processing speed is maintained, but the rarity and impact of dysfunctional pixel malfunctions are overlooked leading to visible defects
Solution Approach 1:
The patent applies local quality by introducing a focused cost function that treats different pixel regions differently. Specifically, pixels identified as potentially aberrant (through attention mechanisms or outlier detection) are assigned higher weight in the cost function, while normal pixels use standard reconstruction costs. This localized weighting ensures that rare dysfunctional pixels receive sufficient attention without compromising overall processing efficiency.
Solution Approach 2:
The patent implements preliminary action by pre-identifying potentially aberrant pixels before final image reconstruction. Through attention mechanisms, outlier detection, or preliminary processing stages, the system flags suspicious pixels in advance. This preliminary identification allows the main reconstruction algorithm to allocate appropriate computational resources and apply specialized correction strategies only where needed, rather than treating all pixels uniformly.
2Reliability
If a focused cost function is applied to correct aberrant pixels, then image quality improves, but computational complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the image processing task into distinct stages: (1) preliminary identification of potentially aberrant pixels through attention mechanisms or outlier detection, (2) application of focused cost function specifically to identified aberrant regions, and (3) standard reconstruction for normal regions. This segmentation allows the complex focused correction to be applied only where necessary, reducing overall computational burden while maintaining high correction effectiveness.
Solution Approach 2:
The patent implements partial action by applying the computationally intensive focused cost function only to a subset of pixels identified as potentially aberrant, rather than processing the entire image with the same level of complexity. This selective application maintains the benefits of focused correction while significantly reducing the overall computational burden compared to applying exhaustive processing to all pixels.
3Measurement precision
If neural network training with modified images is used, then accuracy of dysfunctional pixel detection improves, but training time and computational resources increase
Solution Approach 1:
The patent applies preliminary action in training by pre-generating modified images with known aberrant pixel patterns before actual model training begins. These synthetic training datasets, created through data augmentation techniques that introduce controlled dysfunctional pixel scenarios, allow the neural network to learn from realistic failure modes without requiring extensive collection of real-world defective images, thereby reducing training time while maintaining detection accuracy.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
The present description relates to a method for training a neural network (202) comprising: - generating a modified image, by a modified data generator, on the basis of a first image, the modified image comprising at least one aberrant pixel value with respect to the first image; - providing the modified image to the network; - generating, by the network, a corrected image; - providing the corrected image, by the network, and providing an indication of the position of the at least one modified pixel, by the generator, to a calculation circuit (210) - generating an error value, on the basis of the application of a cost function, by the calculation circuit, taking as input the first image, the indication and the corrected image; - correcting parameters associated with the network by backpropagation of the error in the network.