Imaging Encoder Neural Network for Low-Power HD Data Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As display devices, such as TVs, require higher definition, imaging devices need to produce high-definition image data, which increases the amount of data to be processed, leading to larger circuit sizes and higher power consumption due to the need for more circuits and wirings to transfer this data effectively.
Innovation Solution
An imaging device with an integrated encoder that forms a first neural network for feature extraction using convolution processing and a decoder that forms a second neural network for decompression, utilizing a metal oxide transistor in the channel formation region to reduce circuit area and power consumption by efficiently processing and transferring image data through a hierarchical neural network structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-definition imaging is implemented to meet display device requirements, then image quality is improved, but the amount of image data increases leading to larger circuit area
Solution Approach 1:
The patent extracts and processes only the essential features from high-definition image data through neural network-based feature extraction. Instead of transferring all pixel data, the system identifies and extracts salient features (edges, textures, patterns) that convey the most important visual information, thereby reducing the data volume that needs to be processed by subsequent circuits while maintaining perceived image quality.
Solution Approach 2:
The imaging device divides the high-definition image processing task into multiple stages: initial image capture, feature extraction through neural networks, selective data processing, and output generation. This segmentation allows different parts of the circuit to handle only the data relevant to their function, reducing overall circuit complexity and area requirements while maintaining high-definition output capability.
2Productivity
If more circuits and wirings are added to transfer high-definition image data, then data transfer capability is improved, but device size increases
Solution Approach 1:
The system extracts only the essential feature data from complete high-definition images before transmission and processing. By removing redundant pixel information and keeping only salient features, the data transfer requirement is reduced significantly, allowing high-definition processing capability with fewer and smaller wiring connections.
Solution Approach 2:
The patent transforms the data representation from two-dimensional pixel arrays to a more compact feature-space representation through neural network processing. This dimensional transformation compresses the data structure, reducing the number of data paths and connections needed while maintaining the ability to reconstruct high-definition images at the output stage.
3Measurement precision
If high-definition image data is processed with conventional methods, then image quality is maintained, but power consumption increases
Solution Approach 1:
The system extracts and processes only essential features from high-definition images rather than all pixel data. This selective processing approach maintains image quality by preserving critical visual information while discarding redundant data, thereby reducing the computational load and power consumption of processing circuits.
Solution Approach 2:
The neural network performs feature extraction and data compression before the main image processing and transmission stages. This preliminary action prepares the data in an optimized format that requires less power for subsequent processing, while still enabling high-definition output when needed.
Data Source
AI summary
An imaging device with reduced power consumption is provided.The imaging device includes an imaging portion and an encoder. First image data obtained by the imaging portion is transmitted to the encoder. The encoder includes a first circuit that forms a neural network, and the first circuit conducts feature extraction by the neural network on a first image to generate second image data. Note that since the first circuit has a function of performing convolution processing using a weight filter, the encoder can perform computation with a convolutional neural network.