Bayesian color constancy for outdoor object recognition

Outdoor scene classification is challenging due to irregular geometry, uncontrolled illumination, and noisy reflectance distributions. This paper discusses a Bayesian approach to classifying a color image of an outdoor scene. A likelihood model factors in the physics of the image formation process, sensor noise distribution, and prior distributions over geometry, material types, and illuminant spectrum parameters. These prior distributions are learned through a training process that uses color observations of planar scene patches over time. An iterative linear algorithm estimates the maximum likelihood reflectance, spectrum, geometry, and object class labels for a new image. Experiments on images taken by outdoor surveillance cameras classify known material types and shadow regions correctly, and flag as outliers material types that were not seen previously.

[1]  D. B. Judd,et al.  Spectral Distribution of Typical Daylight as a Function of Correlated Color Temperature , 1964 .

[2]  J. Cohen.  Dependency of the spectral reflectance curves of the Munsell color chips , 1964 .

[3]  J. Pokorny,et al.  Spectral sensitivity of the foveal cone photopigments between 400 and 500 nm , 1975, Vision Research.

[4]  A. E. McGarity,et al.  Solar Engineering Technology , 1985 .

[5]  L. Maloney,et al.  Color constancy: a method for recovering surface spectral reflectance. , 1986, Journal of the Optical Society of America. A, Optics and image science.

[6]  L. Maloney.  Evaluation of linear models of surface spectral reflectance with small numbers of parameters. , 1986, Journal of the Optical Society of America. A, Optics and image science.

[7]  Brian A. Wandell,et al.  The Synthesis and Analysis of Color Images , 1992, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[8]  Yuichi Ohta,et al.  An approach to color constancy using multiple images , 1990, [1990] Proceedings Third International Conference on Computer Vision.

[9]  Glenn Healey,et al.  Segmenting images using normalized color , 1992, IEEE Trans. Syst. Man Cybern..

[10]  Mark S. Drew,et al.  Diagonal transforms suffice for color constancy , 1993, 1993 (4th) International Conference on Computer Vision.

[11]  M. D'Zmura,et al.  Color constancy. I. Basic theory of two-stage linear recovery of spectral descriptions for lights and surfaces. , 1993, Journal of the Optical Society of America. A, Optics, image science, and vision.

[12]  Thomas G. Dietterich.  What is machine learning? , 2020, Archives of Disease in Childhood.

[13]  D H Brainard,et al.  Bayesian color constancy. , 1997, Journal of the Optical Society of America. A, Optics, image science, and vision.

[14]  Brian V. Funt,et al.  Is Machine Colour Constancy Good Enough? , 1998, ECCV.

[15]  Glenn Healey,et al.  What is the spectral dimensionality of illumination functions in outdoor scenes? , 1998, Proceedings. 1998 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (Cat. No.98CB36231).

[16]  Bruce A. Draper,et al.  Color recognition in outdoor images , 1998, Sixth International Conference on Computer Vision (IEEE Cat. No.98CH36271).

[17]  David A. Forsyth,et al.  Sampling, resampling and colour constancy , 1999, Proceedings. 1999 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (Cat. No PR00149).

[18]  Color constancy using KL-divergence , 2001, Proceedings Eighth IEEE International Conference on Computer Vision. ICCV 2001.

[19]  Takeo Kanade,et al.  Statistical Calibration of the CCD Imaging Process , 2001, ICCV.

[20]  S. J.P..  Characteristic spectra of Munsell colors , 2002 .