Deep Neural Network Image Compression and Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a need to prevent deterioration in coding efficiency when the amount of information of an original image to be encoded is increased, and to improve efficiency in the process of receiving a bitstream of an encoded image and decoding the encoded image.
Innovation Solution
An image reconstructing method and device that uses a deep convolutional neural network to perform upsampling and downsampling, allowing for efficient encoding and decoding by managing data throughput and utilizing different filter kernels for upsampling and downsampling processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the amount of information of an original image is increased to improve image quality, then image quality is improved, but coding efficiency deteriorates
Solution Approach 1:
The patent segments the image processing into two distinct stages: first, downsampling to reduce the amount of information and improve coding efficiency; second, upsampling using a deep convolutional neural network to restore image quality. This segmentation allows each stage to be optimized independently, resolving the contradiction between image quality and coding efficiency.
Solution Approach 2:
The patent performs downsampling as a preliminary action before encoding. By reducing the image resolution beforehand, the amount of information to be encoded is decreased, which improves coding efficiency while the subsequent upsampling ensures image quality is maintained.
2Loss of information
If data throughput is increased to maintain image detail, then image detail is preserved, but encoding and decoding efficiency deteriorates
Solution Approach 1:
The patent dynamically adjusts the data throughput at different stages of processing. Downsampling reduces data throughput for efficient encoding, while the deep convolutional neural network performs upsampling to restore image details, effectively managing data flow to optimize both efficiency and detail preservation.
3Measurement precision
If high resolution image data is processed directly, then image quality is maintained, but processing complexity and time increase
Solution Approach 1:
The processing is segmented into downsampling followed by selective upsampling using a deep convolutional neural network only where needed. This segmentation reduces overall processing complexity while maintaining image quality in the final output.
Data Source
Figure 1A~1B
Figure 2A
Figure 2B
AI summary
Provided are an image compressing method including determining a compressed image by performing downsampling using a deep neural network (DNN) on an image; determining a prediction signal by performing prediction based on the compressed image; determining a residual signal based on the compressed image and the prediction signal; and generating a bitstream comprising information about the residual signal, wherein the DNN has a network structure that is predetermined according to training of a downsampling process using information generated in an upsampling process, and an image compressing device for performing the image compressing method. Also, provided are an image reconstructing method of reconstructing a compressed image by using a DNN for upsampling, the compressed image having been compressed by the image compressing method, and an image reconstructing device for performing the image reconstructing method.