However, one of the fundamental weaknesses of the image morphing technique is that intermediate images often have physically incorrect shading such as highlights and shadows. Moreover, we cannot create intermediate images under different viewing and lighting conditions. This is because those images are created by simply interpolating pixel intensities of the two 2D images without knowledge of viewing and illumination conditions and 3D object shapes. In this context, 3D shape morphing techniques have a definite advantage in that arbitrary viewing and illumination conditions can be used for creating new images. Unfortunately, those 3D morphing techniques do not account for object surface reflectance properties or reflection models for generating new images. This often results in undesired shading artifacts (e.g., shadows and highlights fail to match shape changes occurring in the morph).
In this work, we consider a new approach for 3D shape and reflectance morphing of two real 3D objects. Our morphing method consists of two components: shape and reflectance property measurement, and smooth interpolation of the measured properties. For measuring shape and reflectance parameters, the method proposed by Sato and Ikeuchi [2] is used in our analysis. First, a range image and a color image sequence are obtained for each of the two objects while the objects are rotated. Then, for each range image pixel, the sequence of observed colors of a surface point corresponding to the range image pixel is determined. The observed color sequence is separated into the diffuse and specular reflection components by the algorithm used originally by Sato and Ikeuchi [1]. Subsequently, parameters of a reflection function used in our analysis are estimated reliably for the separated reflection components. The reflection model used here is a simplified Torrance-Sparrow model. The estimated reflectance parameters of each surface point are combined with its measured (X,Y,Z) location, and then stored in a new 2D image. Therefore, each pixel in the 2D image has reflectance parameters as well as a (X,Y,Z) location. The 2D image containing the (X,Y,Z) location and reflectance parameters is referred to as an object image in this work. After object images are created for the two objects, the two object images are used to create a new intermediate object image by interpolating the shape and reflectance properties of the two object images. Pixelwise correspondences between the two object images used for the interpolation is established from correspondence of manually drawn line pairs in the two object images. Finally, the newly created intermediate object image is used to render new images of the object under arbitrary viewing and illumination conditions. Those images successfully represent smooth change of surface material type as well as correct shading such as highlights.