Fractal Visual Search for Similar Image Matching at Lower Compute
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing visual search technologies struggle with accuracy and computational efficiency, particularly when searching for similar but not identical images, and are not well-suited for applications requiring dissimilar search results.
Innovation Solution
A system that leverages fractal transforms to group similar images together and uses a combination of deep and shallow convolutional neural networks to perform visual searches, employing a modified image triplet technique and K-nearest neighbors algorithm to identify and rank similar or dissimilar images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning techniques are used for visual search, then accuracy is improved, but computing power requirements increase
Solution Approach 1:
The patent segments the visual search process into two distinct stages: a fractal-based pre-processing stage for image grouping and a deep learning-based post-processing stage for similarity matching. This segmentation allows the computationally intensive deep learning models to operate only on a reduced subset of images after fractal grouping, thereby maintaining high accuracy while reducing overall computing power requirements.
Solution Approach 2:
The patent applies fractal transforms as a preliminary action before deep learning processing. By pre-grouping images into clusters based on fractal similarity, the system prepares the data in advance to reduce the search space. This preliminary action ensures that when deep learning models process images, they are working with a smaller, pre-organized dataset, thus lowering computational burden while preserving search accuracy.
2Measurement precision
If existing visual search techniques are used, then exact matches can be found, but similar but different images cannot be effectively identified
Solution Approach 1:
The patent introduces a similarity threshold parameter that allows the system to differentiate between exact matches and similar but different images. By adjusting this threshold, the system can adapt its behavior to match the specific needs of different applications. This local quality control enables the system to maintain exact match accuracy when needed while also being versatile enough to identify similar but different images when required.
Solution Approach 2:
The patent makes the search criteria dynamic by allowing adjustment of the similarity threshold and other parameters based on the specific application requirements. This dynamic adaptability enables the same system to serve multiple purposes: finding exact matches when the threshold is high, or identifying similar but different images when the threshold is lowered, thus resolving the contradiction between exact match accuracy and adaptability.
3Speed
If fractal transforms are applied to all images, then search speed is improved, but computational resources are consumed
Solution Approach 1:
The patent applies fractal transforms partially - only to the query image and only to images that pass the initial fractal similarity threshold. Rather than processing all images in the database with fractal transforms, the system selectively applies them only where necessary, thereby improving search speed for relevant results while reducing overall computational resource consumption.
Data Source
AI summary
The present disclosure provides a visual search engine (VSE) configured to perform visual searches. The VSE may receive a query image for searching through a plurality of images stored in a dataset. Different ones of the images stored in the dataset may be indexed or logically grouped together based on a fractal transform that associates images depicting similar content with each other. A fractal transform of the query image may be used to identify a plurality of images from the dataset to be searched based on the query image. A modified image triplet technique using the query image, a derived set of similar images, and a derived set of dissimilar images may be utilized to identify features of the images being searched. Search results logic of the VSE may apply deep learning techniques to the feature sets to identify a set of search results to return for the query image.