Local Feature Image Compression via SIFT Descriptors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image compression schemes face limitations in efficiency and complexity due to the lack of advanced edge and segmentation detection tools, leading to suboptimal compression and reconstruction of images.
Innovation Solution
The use of local feature descriptors, such as Scale Invariant Feature Transform (SIFT) descriptors, to generate compressed image data by combining visual and differential feature descriptors, allowing for efficient compression and reconstruction of images without increasing computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If techniques such as increasing the number of prediction directions, partitions and transforms are used, then compression coding efficiency is increased, but the complexity of image encoders and decoders significantly increases
Solution Approach 1:
The patent extracts and utilizes local feature descriptors (such as SIFT descriptors) from images to represent key visual information. By focusing on extracting and encoding only the essential local features rather than processing the entire image with complex transforms, the method achieves compression while avoiding the complexity increase associated with traditional approaches that use multiple prediction directions and partitions.
2Productivity
If edge-based and segmentation-based coding schemes are used, then visual redundancy is taken into account, but the lack of development in edge and segmentation detection tools limits the efficacy and efficiency
Solution Approach 1:
The patent uses pre-computed local feature descriptors (such as SIFT descriptors) that have been extensively developed and optimized in computer vision literature. Instead of developing new edge and segmentation detection tools, the method copies and leverages these established descriptors, which already capture edge and segmentation information effectively. This approach bypasses the limitation of underdeveloped detection tools by using mature, proven descriptors.
3Adaptability or versatility
If local feature descriptors are used to characterize image regions, then scale and rotation invariance is achieved, but additional computational steps are required for descriptor extraction and matching
Solution Approach 1:
The patent performs local feature descriptor extraction as a preliminary step before compression and transmission. By extracting descriptors such as SIFT features in advance and using them to guide the compression process, the method achieves scale and rotation invariance. The descriptors are computed once and then used for both compression and reconstruction, avoiding the need for repeated complex computations during encoding and decoding operations.
Data Source
AI summary
The use of local feature descriptors of an image to generate compressed image data and reconstruct the image using image patches that are external to the image based on the compressed image data may increase image compression efficiency. A down-sampled version of the image is initially compressed to produce an encoded visual descriptor. The local feature descriptors of the image and the encoded visual descriptor are then obtained. A set of differential feature descriptors are subsequently determined based on the differences between the local feature descriptors of the input image and the encoded visual descriptor. At least some of the differential feature descriptors are compressed to produce encoded feature descriptors, which are then combined with the encoded visual feature descriptor produce image data. The image data may be used to select image patches from an image database to reconstruct the image.