To date, Object Data Analysis is the most inclusive type of data analysis, as far as the object spaces are concerned. It extends multivariate data analysis, landmark based shape analysis, and in the infinite dimensional case, it extends functional data analysis. For more details, see Patrangenaru and Ellingson (2015). Imaging data, especially electronic imaging data is the most common type of data of our times. In Statistics, often times, an object is a certain feature extracted from raw imaging data. Big data of complex type, including 2D shapes or projective shapes of 3D scenes, or textured surfaces can be today extracted from digital camera images, using a synergy of Statistics, Computational Algorithms and Geometry, to retrieve such object data. In Statistics, objects are represented as points on metric spaces called object spaces, that are not linear. Object spaces, such as Kendall shape spaces (see Kendall (1984)), affine shape spaces (see Patrangenaru and Mardia 2005), or projective shape spaces (see Patrangenaru et al.(2010)) are often times compact, a property that is not shared by the good old numerical or functional spaces in multivariate or functional data analysis. Typically an object space $(\mathcal M, \rho,\mathcal B_\rho),$ where $\mathcal B_\rho$ is the Borel $\sigma$-algebra generated by open sets of $(M,\rho),$ should have some level of smoothness, to make estimation of location parameters possible, the ideal structure being that of a smooth manifold, with as much symmetry as possible (see Patrangenaru and Ellingson (2015)\cite{PaEl:2015}). A case in points is that of object spaces that are smooth manifolds; for example $P\Sigma_2^5$ can be identified with a real projective plane $\mathbb RP^2.$ Recall that a manifold is a metric space that is locally homeomorphic to an Euclidean space. Unlike in multivariate data analysis, where the object space is $\mathbb R^m,$ for some fixed dimension $m,$ in object data analysis, we have no linear structure on the object space. Therefore the regression model $y_i = F(x_i)+ \varepsilon_i$ does not make sense , when applied to a manifold valued response situation, because $y_i - F(x_i)$ is not well defined operation. Therefore a new approach is needed to address regression with a response $Y$ on an object space. We are also concerned with attempt to study quantitatively perceived color. Ideally, our research would lead to 3D reconstruction of a surface, based on its digital camera colored images.
