Image Processing Network Module Bit Depth Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning technologies face challenges in mobile devices with low computational power, as they require extensive computations, leading to increased processing time and potential image quality degradation when applied to mobile devices.
Innovation Solution
An image processing network module that includes an encoder to reduce the bit depth of input image data, a quantization network to further process the reduced data, and a decoder to restore the bit depth, thereby reducing data processing while maintaining image characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deep learning technology is applied to mobile devices, then image processing capability is improved, but processing time increases and computational power requirements increase
Solution Approach 1:
The patent changes the bit depth parameter of image data from full precision (e.g., 10 bits) to reduced precision (e.g., 6-8 bits) through the encoder, processes this compressed data through the neural network, and then restores the bit depth through the decoder. This parameter transformation enables deep learning to run efficiently on mobile devices with limited computational power while maintaining acceptable processing speeds.
Solution Approach 2:
The patent introduces an intermediary quantization process between the encoder and decoder that uses a quantization network to convert reduced-bit image data back to full precision. This intermediary mechanism allows the system to process data at lower bit depths for speed while restoring quality for output, resolving the contradiction between processing speed and image quality.
2Loss of time
If network complexity is reduced to speed up processing, then processing time decreases, but image quality deteriorates
Solution Approach 1:
The patent segments the image processing task into three distinct stages: encoding (bit depth reduction), neural network processing, and decoding (bit depth restoration). This segmentation allows each stage to be optimized independently - the encoder and decoder can use simple operations to minimize computational complexity while the neural network processes reduced-bit data faster, thus reducing overall processing time without sacrificing image quality.
Solution Approach 2:
By dynamically changing the bit depth parameter during processing, the system can use lower precision (6-8 bits) during neural network computation to reduce computational complexity and processing time, then restore full precision through the decoder. This parameter transformation allows the system to achieve both fast processing and high image quality.
3Use of energy by moving object
If bit depth of image data is reduced, then amount of data processed decreases and power consumption reduces, but image quality may deteriorate
Solution Approach 1:
The patent maintains continuity of useful action by implementing a complete bit depth restoration pipeline. The encoder reduces bit depth to save energy during processing, the neural network operates on this compressed data with reduced power consumption, and the decoder continuously restores the full bit depth to maintain image quality. This continuous process ensures that energy savings are achieved without permanent quality loss.
Solution Approach 2:
The system changes the bit depth parameter dynamically - reducing it during processing to minimize power consumption and then restoring it to maintain image quality. This parameter transformation allows the system to operate at lower power levels while preserving the quality benefits of high-bit-depth image data.
Data Source
AI summary
An image processing network module, an image processing device, and a method of operating the image processing device are provided. The image processing network module includes an encoder configured to receive input image data and change a bit depth of the input image data to generate first image data, a quantization network configured to quantize the first image data to generate second image data, and a decoder configured to receive the second image data and change a bit depth of the second image data to generate output image data.