Search

3D reconstruction based on structured light or laser scan has been widely used in industrial measurement, robot navigation, and virtual reality. However, most modern range sensors fail to scan transparent objects and some other special materials, of which the surface cannot reflect back the accurate depth because of the absorption and refraction of light. In this paper, we fuse the depth and silhouette information from an RGB-D sensor (Kinect v1) to recover the lost surface of transparent objects. Our system is divided into two parts. First, we utilize the zero and wrong depth led by transparent materials from multiple views to search for the 3D region which contains the transparent object. Then, based on shape from silhouette technology, we recover the 3D model by visual hull within these noisy regions. Joint Grabcut segmentation is operated on multiple color images to extract the silhouette. The initial constraint for Grabcut is automatically determined. Experiments validate that our approach can improve the 3D model of transparent object in real-world scene. Our system is time-saving, robust, and without any interactive operation throughout the process. [ABSTRACT FROM AUTHOR]

In recent years, significant progress has been made in detecting text in scene images. However, most of state-of-the-art approaches can not work well when encountered blurred, low-resolution and small-sized texts. We consider many connected regions as candidates, which aim to capture character regions as many as possible. In this paper, we propose a novel method, which is based on exhaustive segmentation, to detect text in scene images. Firstly, we present a parallel structure to generate character candidate regions with the exhaustive segmentation of scene image. Secondly, a well-designed two-layer filtering method is used to filter out non-character candidate regions. Finally, at text line grouping stage, the edges of the fully connected graph of the remaining character candidate regions are cut by a support vector machine classifier. We use two public datasets, namely, ICDAR 2013 dataset and the Street View Text dataset to evaluate the performance of our method. Experimental results show that our method achieves excellent recall rate on these two public datasets, moreover, our method is robust to the blurred, low-resolution and small-sized texts. [ABSTRACT FROM AUTHOR]

Due to the complex backgrounds, size variations, and changes in perspective and orientation in natural scene images, detecting multi-oriented text is a difficult problem that has recently attracted considerable attention from research communities. In this paper, we present a novel method that effectively and robustly detects multi-oriented text in natural scene images. First, the candidate characters are generated by an exhaustive segmentation-based method that can extract characters in arbitrary orientations. Second, a convolutional neural network (CNN) model is employed to filter out the non-character regions; this model is also robust to arbitrary character orientations. Finally, text-line grouping is treated as a problem of pruning non-adjacent graph edges from a graph in which each vertex represents a character candidate region. To evaluate our algorithm, we compare it with other existing algorithms by performing experiments on three public datasets: ICDAR 2013, the Oriented Scene Text Dataset (OSTD) and USTB-SV1K. The results show that the proposed method handles any arbitrary text orientation well, and it achieves promising results on these three public datasets. [ABSTRACT FROM AUTHOR]