Hair detection, segmentation, and hairstyle classification in the wild


Hair detection, segmentation, and hairstyle classification in the wild PDF
U.R. Muhammad*, M. Svanera*, R. Leonardi, and S. Benini*
Image and Vision Computing, 2018.

* These authors contributed equally to this work.

Abstract

Hair highly characterises human appearance. Hair detection in images is useful for many applications, such as face and gender recognition, video surveillance, and hair modelling. We tackle the problem of hair analysis (detection, segmentation, and hairstyle classification) from unconstrained view by relying only on textures, without a-priori information on head shape and location, nor using body-part classifiers. We first build a hair probability map by classifying overlapping patches described by features extracted from a CNN, using Random Forest. Then modelling hair (resp. non-hair) from high (resp. low) probability regions, we segment at pixel level uncertain areas by using LTP features and SVM. For the experiments we extend Figaro, an image database for hair detection to Figaro1k, a new version with more than 1,000 manually annotated images. Achieved segmentation accuracy (around 90%) is superior to known state-of-the-art. Images are eventually classified into hairstyle classes: straight, wavy, curly, kinky, braids, dreadlocks, and short.

Dataset

database
Figaro1k, an extension of Figaro, is a multi-class image database for hair analysis in the wild. It contains 1050 unconstrained view images with persons, subdivided into seven different hairstyles classes (straight, wavy, curly, kinky, braids, dreadlocks, short), where each image is provided with the related manually segmented hair mask.

Proposed Pipeline

workflow
We present the first work for a complete hair analysis in unconstrained setting, including hair detection, segmentation, and hairstyle classification. Hair analysis is performed without using any a-priori information about head and hair location, nor any body-part classifier, but exploiting only texture features. Using deep features derived from CaffeNet-fc7 and RF classification we first tackle hair detection at patch level (a). We then refine the obtained results by segmenting hair at pixel level using LTP feature and SVM classifier (b). As a last step of the processing chain hairstyle is classified by adopting a majority voting scheme on the previously segmented hair patches (c).

Code

Github

BibTeX

@inproceedings{Umar2018Hair,
          title={Hair detection, segmentation, and hairstyle classification in the wild},
          author={Muhammad, Umar Riaz and Svanera, Michele and Benini, Sergio and Leonardi, Riccardo},
          booktitle={Image and Vision Computing},
          year={2018},
          }