Multi-Spectral Image Post-Processing for Object Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multi-spectral image processing methods struggle with accurate object identification due to incorrect pixel classifications, particularly under low light conditions and non-uniform lighting, leading to errors in identifying materials like human skin.
Innovation Solution
A post-processing method that filters disconnected small areas of pixels by calculating density values within defined blocks and comparing them to a threshold, setting all pixels in a block to zero or non-zero values based on the density, and grouping contiguous pixels with the same binary value to form objects, thereby reducing classification errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pixel classification is performed using correlation coefficient method, then most pixels are correctly classified, but many pixels are wrongly classified under low light conditions and non-uniform lighting
Solution Approach 1:
The image is divided into multiple blocks of pixels, and each block is processed independently to determine whether it represents a valid object or noise. This segmentation allows the system to distinguish between genuine objects and classification errors by analyzing the spatial distribution and density of classified pixels within each block.
Solution Approach 2:
The system uses feedback mechanisms by comparing the density of classified pixels in each block against threshold values. The classification results are refined through iterative feedback where blocks failing the density test are re-evaluated or marked as invalid, thereby improving overall classification reliability.
2Productivity
If a low threshold for correlation coefficient is used, then more pixels are classified as objects, but more false positives are introduced
Solution Approach 1:
By segmenting the image into blocks and applying density-based filtering at the block level, the system can tolerate lower individual pixel thresholds while maintaining overall precision. The block-level aggregation compensates for individual false positives through statistical density analysis.
Solution Approach 2:
The system intentionally performs excessive classification by using a low threshold to capture all potential objects, then applies post-processing filtering to remove false positives. This two-stage approach ensures no objects are missed while systematically eliminating errors through density-based block analysis.
3Speed
If pixel classification is performed without post-processing, then processing speed is maintained, but object identification accuracy deteriorates
Solution Approach 1:
The post-processing step segments the classified image into blocks and applies efficient density calculations. This segmented approach processes only relevant regions with computationally simple operations, maintaining high processing speed while significantly improving object identification accuracy through systematic error filtering.
Data Source
AI summary
What is disclosed is a system and method for post-processing a multi-spectral image which has already been processed for pixel classification. A binary image is received which contains pixels that have been classified using a pixel classification method. Each pixel in the image has an associated intensity value and has a pixel value of 1 or 0 depending on whether the pixel has been classified as a material of interest or not. A block of size m×n is defined. Pixel values in a block are changed according to a threshold-based filtering criteria such that pixels in the same block all have the same binary value. The block is then shifted by k pixels and pixel processing repeats until all pixels have been processed. Once all blocks have been processed, contiguous pixels having the same binary value are grouped to form objects. In such a manner, pixel classification errors are reduced.