Wavelet Scattering Compression for High-Volume Parcel Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems face challenges in efficiently compressing and decompressing high-volume image data from items being processed in distribution networks, such as mail and parcels, while maintaining clarity and reducing memory and communication overhead.
Innovation Solution
A system and method utilizing wavelet scattering transform and deep learning to classify and compress images, followed by encoding to remove redundant information, and an inverse process for decompression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional image compression methods are used on high-volume image data from distribution networks, then processing speed is maintained, but compression efficiency and memory usage are insufficient
Solution Approach 1:
The patent segments the image processing task into distinct stages: wavelet scattering transform for feature extraction, deep learning classification for categorization, and targeted compression for storage. This segmentation allows each stage to optimize for its specific function, improving overall compression efficiency while managing memory usage at each step rather than loading entire high-resolution images into memory simultaneously.
Solution Approach 2:
The wavelet scattering transform performs preliminary feature extraction and classification before compression. By pre-processing images to extract essential features and categorize them using deep learning models, the system compresses only the necessary information rather than entire images, significantly reducing memory requirements while maintaining compression efficiency.
2Measurement precision
If deep learning models are trained on large datasets to improve classification accuracy, then model precision increases, but training time and computational resources increase
Solution Approach 1:
The patent applies wavelet scattering transform as a preliminary action before feeding images to deep learning models. This pre-processing step extracts salient features and reduces image dimensionality, allowing models to achieve high classification accuracy with fewer training iterations and less computational time, as the models learn from already-processed feature representations rather than raw pixel data.
3Productivity
If wavelet scattering transform and deep learning are applied to all images, then compression efficiency improves, but processing complexity increases
Solution Approach 1:
The patent applies wavelet scattering transform and deep learning classification selectively based on image characteristics and requirements. Not all images undergo the full processing pipeline - the system adapts the level of processing to local needs, applying complex transformations only where necessary to achieve compression efficiency while avoiding unnecessary processing complexity for simpler cases.
Data Source
AI summary
This application relates to a method and a system for compressing a captured image of an item such as a mailpiece or parcel. The system may include a memory configured to store images of a plurality of items captured while the items are being transported and a processor in data communication with the memory. The processor may be configured to receive or retrieve one or more of the captured images, perform a wavelet scattering transform on the one or more captured images, perform deep learning on the wavelet scattering transformed images to classify the wavelet scattering transformed images and compress the classified wavelet scattering transformed images. Various embodiments can significantly improve a compression efficiency, a communication efficiency of compressed data and save a memory space so that the functionality of computing devices is significantly improved.