Illumination color covariant locale-based visual object retrieval

Abstract

Search by Object Model -finding an object inside a target image -is a desirable and yet difficult mechanism for querying multimedia data. An added difficulty is that objects can be photographed under different lighting conditions. While human vision has color constancy, an invariant processing, presumably, here we seek only covariant processing and look to recover such lighting change. Making use of feature-consistent locales in an image we develop a scene partition by localization, rather than by image segmentation. A diagonal model for illumination change and a voting scheme in chromaticity space provide a candidate set of lighting change coefficients for covariant image transformation. For each pair of coefficients, Elastic Correlation, a form of correlation of locale colors, is performed along with a Least Squares minimization for pose estimation. Since the rotation, scale and translation parameters are thus estimated, we can apply an efficient process of texture support and shape verification. Tests on an image and video database of about 1,500 images show an average Recall and Precision of over 70%.

Shape retrieval with eigen-CSS search

“…The eccentricity e is the ratio of maximum to minimum eigenvalues of the 2 × 2 matrix of second-order central moments of the original, 2D curve, an area-based descriptor. In fact, since the matrix is 2D, this can easily be calculated analytically [15]. Circularity c is the ratio of the square of the perimeter to the area, a contour-based parameter.…”

Online learning of color transformation for interactive object recognition under various lighting conditions

“…Some researchers propose linear color transformation [4] and diagonal color transformation (independent transformation of each RGB channel), which are derived from the physics-based color model. Drew et al [5] propose a voting scheme with pairs of colors of corresponding features to get candidates of diagonal color transformation.…”

“…Miller et al [6] propose a method of non-linear color transformation using color eigenflows learned from multiple pairs of images of a same scene under different lighting conditions. These two methods [5][6] need multiple pairs of reference colors for estimating color transformation. It is however difficult for the robot vision system to get multiple reference colors in unknown lighting conditions.…”

Object Recognition under Various Lighting Conditions

“…Some researchers propose linear color transformation [6] and diagonal color transformation (independent transformation of each RGB channel), which are derived from the physics-based color model. Drew et al [8] propose a voting scheme with pairs of colors of corresponding features to get candidates of diagonal color transformation.…”

“…Miller et al [9] propose a method of non-linear color transformation using color eigenflows learned from multiple pairs of images of a same scene under different lighting conditions. These two methods [8], [9] need multiple pairs of reference colors for estimating color transformation. It is however difficult for the robot vision system to get multiple reference colors in unknown lighting conditions.…”