
This paper presents a method for reconstructing the original high dynamic range (HDR) image from a saturated low dynamic range (LDR) image with missing physical information, specifically for single dielectric objects. A deep neural network approach is employed to map an 8-bit LDR image directly to its corresponding HDR representation. We begin by analyzing the reflection and saturation characteristics of dielectric materials and then construct an HDR image database using a diverse set of dielectric objects. Each HDR image is clipped to generate a set of 8-bit LDR images. All HDR-LDR image pairs are normalized to a fixed resolution and used for training and validation. A deep convolutional neural network (CNN) is designed in the form of an autoencoder architecture with skip connections. The entire network is implemented using MATLAB’s machine learning toolbox, with the ADAM optimizer employed for training. The performance of the proposed method is evaluated using a separate validation set. Comparative experiments with existing methods demonstrate that our approach achieves significantly higher reconstruction accuracy and better histogram fitting.

We describe a comprehensive method for estimating the surface-spectral reflectance from the image data of objects acquired under multiple light sources. This study uses the objects made of an inhomogeneous dielectric material with specular highlights. A spectral camera is used as an imaging system. The overall appearance of objects in a scene results from the chromatic factors such as reflectance and illuminant and the shading terms such as surface geometry and position. We first describe the method of estimating the illuminant spectra of multiple light sources based on detecting highlights appearing on object surfaces. The highlight candidates are detected first, and then some appropriate highlight areas are interactively selected among the candidates. Next, we estimate the spectral reflectance from a wide area selected from an object's surface. The color signals observed from the selected area are described using the estimated illuminant spectra, the surfacespectral reflectance, and the shading terms. This estimation utilizes the fact that the definition domains of reflectance and shading terms are different in each other. We develop an iterative algorithm for estimating the reflectance and the shading terms in two steps repeatedly. Finally, the feasibility of the proposed method is confirmed in an experiment using everyday objects under the illumination environment with multiple light sources.