Discontinuity-preserving decoding of one-shot shape acquisition using regularized color

Abstract The 3D scene acquisition is becoming increasingly crucial in practical application. In this paper, a new method is proposed to measure the surface of objects, which just require one color pattern image. In our decoding method, advancements are made at two steps. First, color identification is modeled as an unsupervised classification problem and K-means are adopted on a new color feature, called regularized color. It is insensitive to surface orientation, illumination direction and illumination intensity for matte, dull surfaces. Second, a discontinuity-preserving method is proposed in the sequence matching, which is based on the window voting to judge correct correspondences and potential borders. In the experiments, this new color feature is compared with RGB, normalized color, HSI. Their class separability measurements are evaluated by scattering criteria and Bhattacharyya distance. The results show that regularized color has much higher discriminating power than RGB and equivalent performance with HSI. Our matching method is also compared with the traditional local matching methods. The results affirm that ours has higher accuracies on six different objects.