Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improving analytical color and texture similarity estimation methods for dataset-agnostic person reidentification

Nikita Gabdullin

TL;DR

The paper tackles the challenge of efficient, dataset-agnostic person re-identification suitable for edge devices by integrating compact human parsing with analytical color and texture features. It introduces color analysis in Lab space with histogram smoothing and a pre-configured latent-space SAE for texture, enabling interpretable, region-wise similarity without reliance on re-id training data. Experimental results on Market1501 show metrics comparable to some deep-learning approaches, though improvements are modest due to limited texture variability, underscoring the need for additional discriminative features. Practically, the approach enables edge deployments and text-based gallery search, while outlining directions for broader texture classes and alternative latent-space configurations to enhance generalization.

Abstract

This paper studies a combined person reidentification (re-id) method that uses human parsing, analytical feature extraction and similarity estimation schemes. One of its prominent features is its low computational requirements so it can be implemented on edge devices. The method allows direct comparison of specific image regions using interpretable features which consist of color and texture channels. It is proposed to analyze and compare colors in CIE-Lab color space using histogram smoothing for noise reduction. A novel pre-configured latent space (LS) supervised autoencoder (SAE) is proposed for texture analysis which encodes input textures as LS points. This allows to obtain more accurate similarity measures compared to simplistic label comparison. The proposed method also does not rely upon photos or other re-id data for training, which makes it completely re-id dataset-agnostic. The viability of the proposed method is verified by computing rank-1, rank-10, and mAP re-id metrics on Market1501 dataset. The results are comparable to those of conventional deep learning methods and the potential ways to further improve the method are discussed.

Improving analytical color and texture similarity estimation methods for dataset-agnostic person reidentification

TL;DR

The paper tackles the challenge of efficient, dataset-agnostic person re-identification suitable for edge devices by integrating compact human parsing with analytical color and texture features. It introduces color analysis in Lab space with histogram smoothing and a pre-configured latent-space SAE for texture, enabling interpretable, region-wise similarity without reliance on re-id training data. Experimental results on Market1501 show metrics comparable to some deep-learning approaches, though improvements are modest due to limited texture variability, underscoring the need for additional discriminative features. Practically, the approach enables edge deployments and text-based gallery search, while outlining directions for broader texture classes and alternative latent-space configurations to enhance generalization.

Abstract

This paper studies a combined person reidentification (re-id) method that uses human parsing, analytical feature extraction and similarity estimation schemes. One of its prominent features is its low computational requirements so it can be implemented on edge devices. The method allows direct comparison of specific image regions using interpretable features which consist of color and texture channels. It is proposed to analyze and compare colors in CIE-Lab color space using histogram smoothing for noise reduction. A novel pre-configured latent space (LS) supervised autoencoder (SAE) is proposed for texture analysis which encodes input textures as LS points. This allows to obtain more accurate similarity measures compared to simplistic label comparison. The proposed method also does not rely upon photos or other re-id data for training, which makes it completely re-id dataset-agnostic. The viability of the proposed method is verified by computing rank-1, rank-10, and mAP re-id metrics on Market1501 dataset. The results are comparable to those of conventional deep learning methods and the potential ways to further improve the method are discussed.

Paper Structure

This paper contains 12 sections, 3 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Loss landscape analysis methodology
  3. Color similarity in Lab color space
  4. Texture similarity with pre-configured LS
  5. Similarity estimation scheme
  6. Experiments
  7. Color histogram modification experiments
  8. Feature channel similarity weight experiments
  9. Parser class weight experiments
  10. Gallery search without query images
  11. Discussions
  12. Conclusions

Figures (5)

  • Figure 1: L channel histogram before and after smoothing; values above the threshold line are used to choose non-zero bins for binary histograms.
  • Figure 2: Visualization of pre-configured LS of SAE with projections of two images and their distances to centers of the closest clusters.
  • Figure 3: Search results for persons wearing "checkered upper clothes" in Market1501. Numbers above images show similarity score out of 8.
  • Figure 4: Search results for persons wearing "white checkered upper clothes" in Market1501. Numbers above images show similarity score out of 8.
  • Figure 5: Search results for persons wearing "red upper clothes and black pants" in Market1501. Numbers above images show similarity score out of 14.