Visual Output Testing With Perceptual Hashing Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated verification techniques struggle to accurately evaluate sensory-based visual output generated by computer software due to imperceptible differences, leading to increased manual review and costs, as bit-wise comparisons often flag images as inaccurate despite being visually indistinguishable.
Innovation Solution
A software development test platform utilizing perceptual hashing, specifically through discrete cosine transform, compares visual output to a reference image at a perceived information level, determining acceptable output based on predefined thresholds for minor imperceptible differences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If bit-wise comparison is used to verify visual output, then automated verification accuracy is improved, but manual review frequency increases due to false positives from imperceptible differences
Solution Approach 1:
The patent transforms the verification parameter from bit-level differences to perceptual hash similarity scores. By converting images to perceptual hash representations and comparing similarity thresholds, the system identifies visually imperceptible differences as acceptable while catching actual visual defects, thereby reducing false positives and manual review frequency while maintaining verification accuracy
Solution Approach 2:
The patent replaces the mechanical bit-by-bit comparison system with a perceptual hashing system that mimics human visual perception. This substitution uses hash similarity scoring to evaluate visual output, aligning automated verification with human perceptual judgment and eliminating the discrepancy between machine detection and human acceptance
2Measurement precision
If test engineers perform manual evaluation of visual output, then verification accuracy is improved by considering perceptual differences, but cost and time consumption increase
Solution Approach 1:
The patent enables the automated verification system to evaluate visual output independently using perceptual hashing algorithms that replicate human perceptual judgment. The system autonomously determines whether differences are visually imperceptible without requiring human intervention, achieving both high accuracy and efficiency simultaneously
Solution Approach 2:
The patent introduces perceptual hash representation as an intermediary between the visual output and the verification decision. This intermediary transforms images into a domain where perceptual similarity can be efficiently computed, allowing automated systems to make human-like judgment calls without time-consuming manual review
3Productivity
If bit-wise comparison is used to verify visual output, then automated verification speed is improved, but reliability decreases due to false rejection of perceptually identical images
Solution Approach 1:
The patent changes the verification parameter from exact bit-match to perceptual hash similarity with threshold-based acceptance. This allows the system to maintain high verification speed while improving reliability by accepting images that are perceptually identical despite bit-level variations, thereby reducing false rejections
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A software development test platform, method and computer program product are provided to evaluate, on a perceived information level, visual output generated by execution of computer software. In the context of a method, data is provided that is representative of the visual output generated in response to execution of computer software. The method also includes comparing the data representative of the visual output to a reference image utilizing perceptual hashing. The comparison of the data representative of the visual output to the reference image includes comparing a hash value of the data representative of the visual output and a hash value of the reference image to define a perceptual difference between the visual output and the reference image. The method further includes determining the visual output to be acceptable in an instance in which the perceptual difference satisfies a predefined threshold.