Investigation Of Reflection Component Separation And Spectrum Measurement In Color Imaging

Color imaging technology plays an important role in color science; it provides experimental basis for color research and application. Color information is traditionally represented by red, green and blue values, which are not always sufficient for industrial applications. The multispectral imaging system combines the advantages of both camera and spectrophotometer, and provides a way to measure spectra with a very high spatial resolution. In color measurement, we should deal with the problems in traditional imaging, as well as the particular problems in multispectral imaging. This thesis first investigates the problem of specular highlight removal in color images, and then works on two important issues in multispectral imaging. The main contributions of this dissertation are summarized as follows:1. We propose a novel method to separate the diffuse and specular reflection components from a single image using dichromatic reflection model. We define intensity ratio as the ratio between the maximum value and range value (maximum minus minimum). The intensity ratios of diffuse pixels are independent of surface geometry. The specular components of the pixels can be separated by using the intensity ratios of diffuse pixels with identical body color. For multicolored surfaces, we use pseudo-chromaticity to classify all pixels into different clusters, then treat each cluster as a monochrome region. Experimental results indicate that our method performs better in highlight removal accuracy and4×faster than the state of the art.2. The refractive index of the narrow band filters and camera lens changes with wavelength, and thus the focal length of each channel is different. In multispectral imaging systems, it is necessary to adjust the lens position of each channel to obtain sharp images. This thesis presents a simple and efficient autofocus algorithm based on the symmetry of focus measure. We capture a number of images at equidistant lens positions and simulate the focus measure distribution using spline interpolation. The optimal lens position can be estimated by maximizing the symmetry of the focus measure distribution. We utilize similarity measure to evaluate the symmetry of the focus measure curve. Our method performs well in multispectral imaging system and it can be used in other relevant optical imaging system.3. Due to the different geometries and principles between multispectral imaging system and spectrophotometer, the accuracy of spectral reflectance estimation usually degrades when the material of test samples is different from that of calibration samples. In this thesis, we introduce a novel method to improve the inter-material agreement in spectrum measurement by using adaptive band regression. This band selection problem is casted as a binary optimization problem, which is solved using a novel binary differential evolution algorithm. The experimental results show that our method can yield improved correction accuracy when compared with traditional correction methods.