Image Quality Assessment for Computer Vision Based Perception Algorithms Using Edge and Structural Features

Image Quality Assessment (IQA) is a critical task in image processing, computer vision, and other related fields, as it helps in evaluating the effects of various impairments on the Quality of Experience (QoE) of consumers. However, current IQA metrics may have limitations in evaluating image qualit...

Full description

Saved in:
Bibliographic Details
Published in:TENCON ... IEEE Region Ten Conference pp. 158 - 163
Main Authors: Akshaya, V G, S, Rithika, A, Lanciya Angel, Amirtha Pravin, K S M, Balasubramanian, Karthi, Srinivasan, Suganthi, Rajegowda, Rakesh, Bhattacherjee, Payal
Format: Conference Proceeding
Language:English
Published: IEEE 31.10.2023
Subjects:
BRISQUE
Computer vision
Correlation
CPBD
Image edge detection
Image quality
IQ analysis
IQ metric
JNB
Measurement
NLIEE
Object detection
Perception
Visualization
VQA
ISSN:2159-3450
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Image Quality Assessment (IQA) is a critical task in image processing, computer vision, and other related fields, as it helps in evaluating the effects of various impairments on the Quality of Experience (QoE) of consumers. However, current IQA metrics may have limitations in evaluating image quality accurately for different types of images or under different conditions. To address this, a proposed approach involves calculating an overall image quality score based on linear combination of individual image quality metrics. This method considers multiple image features such as sharpness, contrast, color, and noise, and assigns weights to each feature based on its relative importance in determining overall image quality. By combining and weighting multiple image features, the proposed approach aims to provide a more comprehensive and accurate evaluation of image quality compared to using individual metrics alone. In this study, we focus on the importance of edge features and structure-based metrics in object detection and analyze existing No-Reference (NR) IQA metrics, including Just Noticeable Blur (JNB), Cumulative Probability Blur Detection (CPBD), Visual Quality Assessment (VQA), Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) and No-Reference Low- Light Image Enhancement Evaluation (NLIEE), and propose a linear combination formula that combines these metrics to evaluate image. We evaluate the performance of the proposed metric on several datasets by calculating the linear combination of two, three, and all the five metrics. The weights assigned to each metric in the linear combination formula are determined through experimental analysis. The combined metric scores and MOS scores are then used to compute the Spearman's rank order correlation coefficient, which measures the monotonic relationship between two variables. A higher SROCC value indicates a stronger correlation between the combined metric and MOS scores and is used to evaluate the performance of the metric. Our results show that the proposed metric, which combines VQA, BRISQUE and JNB, provides the highest correlation with MOS scores, over all the datasets. This suggests that the proposed approach can provide an effective and comprehensive way of evaluating image quality in object detection algorithms. Our findings can have significant implications in improving the overall performance of object detection algorithms and enhancing the QoE of consumers.
ISSN:2159-3450
DOI:10.1109/TENCON58879.2023.10322345