Image Signal Processor Neural Network Sub-Pixel Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image sensors placed under display panels in electronic devices often produce deteriorated image quality due to light distortion, leading to issues like flare, haze, and blur, which existing technologies have not adequately addressed.
Innovation Solution
An image signal processor utilizing neural networks to correct global pixel values by splitting and processing sub-pixel values, and sub-feature values, enhancing image quality by correcting distortions such as flare and haze, and improving product reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If an image sensor is disposed under a display panel to enable full-screen sensing, then the sensing area and integration are improved, but image quality deteriorates due to light distortion, flare, haze, and blur
Solution Approach 1:
The patent divides the image signal processing into multiple stages: first splitting the image signal into multiple channels, then applying different neural network correction models to each channel. This segmentation allows targeted correction of different types of light distortion (flare, haze, blur) affecting different parts of the image, thereby maintaining high image quality across the full sensing area.
Solution Approach 2:
The patent introduces neural network-based correction models as intermediary processing layers between the image sensor and final image output. These correction models act as mediators that specifically address light distortion issues caused by the display panel, transforming degraded input signals into corrected image data without requiring physical modification of the sensor-display arrangement.
2Reliability
If neural network-based correction is applied to the entire image signal, then image quality is improved, but processing complexity and computational load increase
Solution Approach 1:
The patent segments the image signal into multiple channels and applies different correction models to each channel based on its specific characteristics. This approach reduces overall processing complexity by avoiding the application of a single complex model to the entire image, instead using multiple simpler, specialized models that can be processed in parallel.
Solution Approach 2:
The patent applies different correction strategies to different parts of the image signal based on local characteristics. By identifying and correcting specific distortion types (flare, haze, blur) in different regions or channels, the system achieves high image quality without uniformly applying complex processing across the entire image, thereby optimizing computational efficiency.
Data Source
AI summary
An image processing device including a memory; and at least one image signal processor configured to: generate, using a first neural network, a feature value indicating whether to correct a global pixel value sensed during a unit frame interval, and generate a feature signal including the feature value; generate an image signal by merging the global pixel value with the feature signal; split a pixel value included in the image signal into a first sub-pixel value and a second sub-pixel value, split a frame feature signal included in the image signal into a first sub-feature value corresponding to the first sub-pixel value and a second sub-feature value corresponding to the second sub-pixel value, and generate a first sub-image signal including the first sub-pixel value and the first sub-feature value, and a second sub-image signal including the second sub-pixel value and the second sub-feature value; and sequentially correct the first sub-image signal and the second sub-image signal using a second neural network.