Combined Machine Learning with Multi-view Modeling for Robust Wound Tissue Assessment

Hazem Wannous (1) , Yves Lucas (1) , Sylvie Treuillet (1)

1 IRAUS - Département Images, Robotique, Automatique et Signal [Orléans]

From colour images acquired with a hand held digital camera, an innovative tool for assessing chronic wounds has been developed. It combines both types of assessment, colour analysis and dimensional measurement of injured tissues in a user-friendly system. Colour and texture descriptors have been extracted and selected from a sample database of wound tissues, before the learning stage of a support vector machine classifier with perceptron kernel on four categories of tissues. Relying on a triangulated 3D model captured using uncalibrated vision techniques applied on a stereoscopic image pair, a fusion algorithm elaborates new tissue labels on each model triangle from each view. The results of 2D classification are merged and directly mapped on the mesh surface of the 3D wound model. The result is a significative improvement in the robustness of the classification. Real tissue areas can be computed by retro projection of identified regions on the 3D model.

Mots clés: Tissue classification; 3D modeling; machine learning; wound assessment

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]

Hazem Wannous, Yves Lucas, Sylvie Treuillet. Combined Machine Learning with Multi-view Modeling for Robust Wound Tissue Assessment. VISAPP 2010 - Fifth International Conference on Computer Vision Theory and Applications, May 2010, Angers, France. pp.92-104. ⟨hal-00648511⟩