JPEG Classifier Using Partial Decompression for Efficient Image Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for classifying compressed images, such as JPEG images, require full decompression followed by feature extraction and classification, which are computationally expensive and power-intensive, especially in devices with limited resources.
Innovation Solution
A method that involves partial decompression to obtain block coefficients from compressed images, which are then used to apply a classifier trained on decoded coefficients, allowing for efficient rough classification with minimal computational overhead. If the rough classification indicates the presence of an object of interest, the image undergoes full decompression and classification for validation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full decompression is performed before classification, then classification accuracy is improved, but computational cost and power consumption increase significantly
Solution Approach 1:
The patent applies partial decompression to obtain only the necessary block coefficients for classification without performing full decompression. This selective approach retrieves sufficient information for accurate classification while avoiding the computational overhead of complete decompression, thus resolving the contradiction between accuracy and power consumption
Solution Approach 2:
The classification process is segmented into two stages: a rough classification stage using partially decompressed block coefficients, and a full classification stage using fully decompressed image data only when necessary. This segmentation allows the system to use minimal computational resources for initial filtering while reserving full processing power only when needed, addressing the energy-accuracy tradeoff
2Reliability
If full decompression and classification are performed on all images, then classification reliability is improved, but processing time increases significantly
Solution Approach 1:
The patent performs preliminary rough classification using partially decompressed block coefficients before committing to full decompression and classification. This preliminary action filters out images that can be confidently classified without full processing, reducing overall processing time while maintaining reliability through a two-stage verification system
Solution Approach 2:
The classification process dynamically adapts its complexity based on the image content and initial classification results. Images requiring high reliability undergo full decompression and classification, while images that can be confidently classified through partial processing skip the expensive full decompression step, optimizing processing time without sacrificing necessary reliability
3Productivity
If partial decompression is used for classification, then computational cost is reduced, but classification accuracy may deteriorate
Solution Approach 1:
The patent introduces block coefficients as an intermediary representation that bridges the compressed image and the classification system. These coefficients serve as a sufficient statistic that captures essential image features for classification without requiring full decompression, enabling both high efficiency and maintained accuracy through the intermediary block coefficient representation
4Productivity
If a two-stage classification system is implemented, then processing efficiency is improved, but system complexity increases
Solution Approach 1:
The block coefficient extraction and classification modules serve multiple functions: they enable rough classification, provide features for full classification when needed, and can operate independently for efficiency. This multi-functionality reduces the need for separate dedicated components, managing system complexity while achieving the benefits of two-stage processing
Data Source
AI summary
A method, device and computer program product, the method comprising: obtaining access to a classifier trained upon a multiplicity of sets of decoded coefficients; obtaining a set of block coefficients associated with at least a part of the compressed image; and applying the classifier to the set of block coefficients, to obtain a classification of the compressed image.


