Automatic Body Feature Extraction from Front and Side Images

Lingyan Jiang, Jian Yao, Baopu Li, Fei Fang, Qi Zhang, Max Q.‐h. MengJournal of Software Engineering and Applications 2012 · DOI

Abstract

Human body feature extraction based on 2D images provides an efficient method for many applications, such as non-contact body size measurements, constructing 3D human models, and recognizing human actions. In this paper, a systematic approach is proposed to detect feature points of the human body automatically from its front and side images. Firstly, an efficient approach for silhouette and contour detection is used to represent the contour curves of a human body shape with Freeman’s 8-connected chain codes. The contour curves are considered as a number of segments connected together. Then, a series of feature points on the human body are extracted based on specified rules by measuring the differences between the directions of the segments. In total, 101 feature points with clear geometric properties are extracted automatically, including 27 points corresponding to the definitions of landmarks for garment measurements. Finally, the proposed approach was tested on ten human subjects, and all 101 feature points were correctly extracted, indicating an effective and robust performance.

Introduction

Traditional anthropometry, or the measurement of the human body, is often a manual, tedious, and subjective process. While 3D body scanners offer a high-tech alternative, their high cost and the complexity of processing 3D point clouds limit their accessibility. Previous attempts to automate feature extraction from 2D images frequently suffered from "cascading errors," where the misidentification of a primary landmark (like the top of the head) led to inaccuracies in all subsequent dependent points.

This paper addresses these issues by proposing a robust system that identifies feature points independently based on the local geometry of the body's contour. This approach makes the system more reliable for applications in the garment industry and virtual reality.

Methodology

The proposed system utilizes a multi-stage pipeline to transform standard photographs into a map of anatomical landmarks:

  1. Image Pre-processing: Front and side images are captured against a plain background. The system employs background subtraction and Canny edge detection to isolate the human silhouette.
  2. Contour Representation: The silhouette is converted into Freeman’s 8-connected chain codes. This represents the boundary as a sequence of directional changes (0–7), where each number corresponds to a 45° shift.
  3. Segment Analysis: The contour is decomposed into segments, and the "inclination" or direction of each segment is analyzed.
  4. Rule-Based Extraction: The core innovation lies in 16 specific geometric rules defined by the relationship between three adjacent segments. By measuring directional differences—identifying where a curve "bumps" out or turns inward—the algorithm identifies feature points locally.
Front view image (a) used for silhouette and feature extraction.

Experimental Setup and Results

The study utilized a custom dataset of ten human subjects (5 males and 5 females). Subjects wore light clothing and headgear to minimize hair interference and stood in a standardized posture with arms slightly away from the body and fingers spread. Images were captured using a standard digital camera.

The algorithm successfully extracted all 101 feature points (71 from the front view and 30 from the side view) across all test subjects. Visual inspection confirmed that these points accurately reflected anatomical "bumps" and "turns," such as the earlobe, shoulder acromion, armpit, and ankle. Notably, 27 of these points were specifically chosen to correspond with international standards for garment measurement (ISO 8559 and GB/T 16160-2008).

Conclusion and Future Work

By treating each feature point as a result of local contour changes rather than a relative coordinate from a "master" point, the system avoids the propagation of errors found in earlier research. This geometric mapping successfully bridges the gap between abstract 2D silhouettes and standardized medical or tailoring landmarks.

Future work will focus on using these extracted points to calculate actual physical dimensions for non-contact tailoring and integrating the 2D feature points into mesh-generation algorithms to automatically generate accurate 3D virtual avatars.

Summary updated

Abstract

Human body feature extraction based on 2D images provides an efficient method for many applications, e.g. non-contact body size measurements, constructing 3D human model and recognizing human actions. In this paper a systematic approach is proposed to detect feature points of human body automatically from its front and side images. Firstly, an efficient approach for silhouette and contour detection is used to represent the contour curves of a human body shape with Freeman’s 8-connected chain codes. The contour curves are considered as a number of segments connected together. Then, a series of feature points on human body are extracted based on the specified rules by measuring the differences between the directions of the segments. In total, 101 feature points with clearly geometric properties (that rather accurately reflect the bump or turning of the contours) are extracted automatically, including 27 points corresponding to the definitions of the landmarks about garment measurements. Finally, the proposed approach was tested on ten human subjects and the entire 101 feature points with specific geography geometrical characteristics were correctly extracted, indicating an effective and robust performance.
BibTeX
@article{jiang2012automatic,
  author = {Lingyan Jiang and Jian Yao and Baopu Li and Fei Fang and Qi Zhang and Max Q.‐H. Meng},
  title = {{Automatic Body Feature Extraction from Front and Side Images}},
  journal = {Journal of Software Engineering and Applications},
  year = {2012},
  doi = {10.4236/jsea.2012.512b019},
}
Figure