Surface reflectance components separation from single color images using the mean-shift decomposition technique Article

Lahlou, M, Adjouadi, M. (2012). Surface reflectance components separation from single color images using the mean-shift decomposition technique . International Journal of Innovative Computing, Information and Control, 8(7 B), 5149-5164.

cited authors

  • Lahlou, M; Adjouadi, M

abstract

  • This study provides a resolution to the separation of specular and diffuse reflectance components in images of textured scenes. The proposed method can be used to solve several challenging tasks associated with computer vision applications ranging from specularity removal, image filtering, and surface reconstruction. We present a unified framework to achieve object surface reflectance separation by studying the dissimilarities between the reflectance components distribution in scene images delineated on a normalized color space. A simple but robust reflectance decomposition technique is introduced based on the Eigen-decomposition transform we named the Mean-Shift Decomposition (MSD) method. This technique provides a direct access to surface shape information through diffuse shading pixels isolation. In addition, the proposed method does not require any local color segmentation process as it differentiates between both reflectance components efficiently. This is viewed as a significant contribution to the prevailing approach of several proposed methods in the literature that operate on images by aggregating color information along each image plane. To recover objects surface geometry information, we formulate a specularity removal process by shifting the specular reflectance components toward the decomposed diffuse reflectance distribution. An empirical evaluation of the proposed reflectance separation technique is performed on several images comprising uniform color surfaces, multicolor surfaces, and highly textured surfaces. © 2012 ICIC International.

authors

publication date

  • July 1, 2012

published in

start page

  • 5149

end page

  • 5164

volume

  • 8

issue

  • 7 B