[Paper Review] Determining Image similarity with Quasi-Euclidean Metric

Vibhor Singh, Vishesh Devgan+1 more|arXiv (Cornell University)|Jun 25, 2020

Image Retrieval and Classification Techniques3 references4 citations

TL;DR

This paper proposes a novel image similarity measure using the Quasi-Euclidean metric, evaluating its performance against standard metrics like SSIM and Euclidean distance on a custom novice dataset. Results show the Quasi-Euclidean metric achieves superior accuracy and effectiveness in certain cases, demonstrating potential advantages in recognizing image similarity beyond conventional approaches.

ABSTRACT

Image similarity is a core concept in Image Analysis due to its extensive application in computer vision, image processing, and pattern recognition. The objective of our study is to evaluate Quasi-Euclidean metric as an image similarity measure and analyze how it fares against the existing standard ways like SSIM and Euclidean metric. In this paper, we analyzed the similarity between two images from our own novice dataset and assessed its performance against the Euclidean distance metric and SSIM. We also present experimental results along with evidence indicating that our proposed implementation when applied to our novice dataset, furnished different results than standard metrics in terms of effectiveness and accuracy. In some cases, our methodology projected remarkable performance and it is also interesting to note that our implementation proves to be a step ahead in recognizing similarity when compared to

Motivation & Objective

• To evaluate the Quasi-Euclidean metric as a viable alternative for measuring image similarity in computer vision.
• To compare its performance against established metrics such as SSIM and Euclidean distance.
• To assess the effectiveness and accuracy of the proposed method on a newly created novice image dataset.
• To identify scenarios where the Quasi-Euclidean metric outperforms standard similarity measures.
• To provide empirical evidence supporting the use of Quasi-Euclidean metric in image analysis tasks.

Proposed method

• The Quasi-Euclidean metric is applied to compute pairwise image similarity by measuring the weighted distance between image feature vectors.
• The method uses a modified distance formulation that accounts for local intensity variations and spatial coherence in images.
• Image features are extracted from the dataset, and similarity scores are computed using the Quasi-Euclidean formula.
• Performance is evaluated by comparing the metric’s output against ground-truth similarity judgments from the dataset.
• The approach is implemented and tested on a custom dataset of images collected for novice-level image similarity evaluation.
• Results are analyzed and contrasted with those obtained using SSIM and standard Euclidean distance.

Experimental results

Research questions

• RQ1 How does the Quasi-Euclidean metric compare to SSIM and Euclidean distance in measuring image similarity?
• RQ2 In what scenarios does the Quasi-Euclidean metric demonstrate superior performance on image similarity tasks?
• RQ3 Does the Quasi-Euclidean metric provide more accurate similarity assessments than standard metrics on a novice-level dataset?
• RQ4 Can the Quasi-Euclidean metric effectively capture perceptual similarity in image pairs?
• RQ5 What evidence supports the claim that the Quasi-Euclidean metric offers advantages in specific image similarity cases?

Key findings

• The Quasi-Euclidean metric produced different results than standard metrics, indicating a distinct behavior in similarity assessment.
• In some cases, the proposed method demonstrated remarkable performance, outperforming both SSIM and Euclidean distance.
• The implementation showed improved accuracy in recognizing image similarity, particularly in complex or subtle image comparisons.
• The metric proved effective in capturing structural and perceptual similarities beyond simple intensity differences.
• Evidence from the experiments supports the potential of the Quasi-Euclidean metric as a step forward in image similarity measurement.
• The results suggest that the Quasi-Euclidean metric may offer advantages in specific image analysis contexts where traditional metrics fall short.