Block-Based Image Feature Detection for Resource-Constrained Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current feature detection algorithms, such as SIFT, require high computational complexity and memory, limiting their implementation in resource-constrained devices like mobile image-capturing devices, and offer limited parallelism at increasing image resolutions.
Innovation Solution
A block-based method for image feature detection that processes image stripes in the frequency domain, reducing memory requirements and processing time by transforming and filtering block data, allowing for parallel processing and keypoint localization without the need for pixel-domain processing, and generates keypoint descriptors using frequency components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If SIFT algorithm is used for feature detection, then feature detection accuracy and robustness are improved, but computational complexity and memory requirements increase significantly
Solution Approach 1:
The patent divides the image into multiple stripes (horizontal or vertical bands) and processes each stripe independently into blocks. This segmentation reduces the overall computational load by processing smaller portions of the image simultaneously, making the SIFT algorithm feasible for resource-constrained devices while maintaining detection accuracy through systematic coverage of all image regions.
Solution Approach 2:
The patent transforms image data from the spatial domain to the frequency domain using Fast Fourier Transform (FFT). This dimensional transformation enables efficient filtering and keypoint detection through frequency-based operations rather than pixel-domain processing, significantly reducing computational complexity while preserving feature detection robustness.
2Measurement precision
If SIFT algorithm processes entire image, then feature detection completeness is improved, but processing time and memory usage increase
Solution Approach 1:
The patent segments the image into multiple stripes that can be processed in parallel. Each stripe is further divided into blocks, enabling concurrent processing of multiple image regions. This segmentation maintains detection completeness by ensuring all image areas are covered while dramatically reducing processing time through parallel execution.
Solution Approach 2:
The patent performs preliminary frequency transformation and filtering operations on each stripe-block before detailed keypoint analysis. This preliminary processing identifies potential keypoint candidates early, allowing subsequent processing to focus only on relevant regions, thus reducing overall processing time while maintaining completeness.
3Reliability
If SIFT algorithm uses scale and octave buffering, then feature detection accuracy is improved, but memory requirements increase
Solution Approach 1:
The patent segments the image into stripes and blocks, processing only the current stripe-block in memory at any time. This segmentation eliminates the need for large-scale buffering across the entire image, reducing memory requirements while maintaining accuracy through systematic processing of all segments.
Solution Approach 2:
By transforming data to the frequency domain, the patent enables efficient processing with reduced memory footprint. Frequency-domain operations allow scale and octave processing without requiring large spatial buffers, as transformations can be performed in-place or with minimal additional memory for frequency representations.
Data Source
AI summary
An embodiment is a method for detecting image features, the method including extracting a stripe from a digital image, the stripe including of a plurality of blocks; processing the plurality of blocks for localizing one or more keypoints; and detecting one or more image features based on the one or more localized keypoints.