Image Compression Device Using CNN Object Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image compression technologies, such as the H.264 standard, face challenges in achieving high compression rates without degrading image quality, especially in complex environments, due to their high hardware complexity and inefficient processing algorithms.
Innovation Solution
An image compression device utilizing a convolutional neural network (CNN) to separate objects and backgrounds within an image, adjusting quantization parameters based on pixel change amounts to optimize compression, thereby enhancing compression rates and image quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of substance
If H.264 standard is used for image compression, then image compression is achieved, but hardware complexity increases and processing efficiency decreases
Solution Approach 1:
The image is divided into multiple blocks, and each block is independently processed with appropriate quantization parameters. The object extracting unit segments the image into object regions and background regions, allowing different compression strategies to be applied to different segments, thereby reducing overall hardware complexity while maintaining compression efficiency.
Solution Approach 2:
Different quantization parameters are applied to different regions of the image based on their importance. Object regions use smaller quantization parameters to preserve quality, while background regions use larger parameters for higher compression. This local differentiation reduces the need for complex global processing hardware.
2Speed
If HEVC standard is used for high speed processing, then processing speed increases, but hardware complexity increases significantly
Solution Approach 1:
The object extracting unit performs preliminary segmentation of the image into object and background regions before the main compression process. By pre-identifying important regions, the subsequent compression can focus computational resources on these areas, achieving fast processing without requiring complex HEVC hardware for entire image processing.
Solution Approach 2:
Instead of applying complex algorithms to the entire image, the system applies detailed processing only to object regions (partial action), while using simpler compression for background regions. This selective approach achieves high processing speed without the full hardware complexity of standards like HEVC.
3Device complexity
If uniform quantization parameter is applied to entire image, then processing is simplified, but image quality of important regions deteriorates
Solution Approach 1:
The system applies different quantization parameters to different regions: smaller QP values for object regions to preserve quality, and larger QP values for background regions. This local quality differentiation maintains image quality for important content while keeping processing complexity manageable through automated region classification.
4Manufacturing precision
If object-based region adjustment is implemented, then image quality is improved, but processing time increases
Solution Approach 1:
The object extracting unit performs preliminary identification and classification of object regions before the main compression process. By pre-segmenting the image into object and background regions with clear boundaries, the subsequent quality-adjusted compression can proceed efficiently without time-consuming iterative processing, thus improving image quality without excessive processing time.
Data Source
AI summary
Provided is an image compression device including an object extracting unit configured to perform convolution neural network (CNN) training and identify an object from an image received externally, a parameter adjusting unit configured to adjust a quantization parameter of a region in which the identified object is included in the image on the basis of the identified object, and an image compression unit configured to compress the image on the basis of the adjusted quantization parameter.