Min Ki Park

Multi-scale tensor voting for feature extraction from unstructured point clouds

Identifying sharp features in a 3D model is essential for shape analysis, matching and a wide range of geometry processing applications. This paper presents a new method based on the tensor voting theory to extract sharp features from an unstructured point cloud which may contain random noise, outliers and artifacts. Our method first takes the voting tensors at every point using the corresponding neighborhoods and computes the feature weight to infer the local structure via eigenvalue analysis of the tensor. The optimal scale for a point is automatically determined by observing the feature weight variation in order to deal with both a noisy smooth region and a sharp edge. We finally extract the points at sharp features using adaptive thresholding of the feature weight and the feature completion process. The multi-scale tensor voting of a given point set improves noise sensitivity and scale dependency of an input model. We demonstrate the strength of the proposed method in terms of efficiency and robustness by comparing it with other feature detection algorithms.

Local color transfer between images using dominant colors

Abstract. A color-transfer method that can transfer colors of an image to another for the local regions using their dominant colors is proposed. In order to naturally transfer the colors and moods of a target image to a source image, we need to find the local regions of colors that need to be modified in the image. Since the dominant colors of each image can be used for the estimation of color regions, we develop a grid-based mode detection, which can efficiently estimate dominant colors of an image. Based on these dominant colors, our proposed method performs a consistent segmentation of source and target images by using the cost-volume filtering. Through the segmentation procedure, we can estimate complex color characteristics and transfer colors to the local regions of an image. For an intuitive and natural color transfer, region matching is also crucial. Therefore, we use the visually prominent colors in the image to meaningfully connect each segmented region in which modified color-transfer method is applied to balance the overall luminance of the final result. In the Experimental Result section, various convincing results are demonstrated through the proposed method.

Spatial augmented reality for product appearance design evaluation

Abstract Augmented reality based on projection, called “Spatial Augmented Reality (SAR)”, is a new technology that can produce immersive contents by overlapping virtuality and real-world environment. It has been paid attention as the next generation digital contents in media art and human–computer interaction (HCI). In this paper, we present a new methodology to evaluate the product appearance design more intuitively by means of SAR technique. The proposed method first projects the high-quality rendered image considering the optical property of materials onto the mock-up of a product. We also conduct a projector-camera calibration to compensate a color distortion according to a projector, a projection surface and environment lighting. The design evaluation methodology we propose offers more flexible and intuitive evaluation environment to a designer and user (evaluator) than previous methods that are performed via a digital display. At the end of this research, we have conducted a case study for designing and evaluating appearance design of an automobile.

SMI 2013: Feature-aware filtering for point-set surface denoising

In this paper, we propose a simple and effective feature-aware filtering for point-set surface denoising, which can achieve a second-order surface approximation of the underlying surface. Our method consists of two stages: robust normal estimation considering sharp features and feature-preserving denoising using a local curvature based projection. The normal clustering based on neighborhood grouping allows the filter to preserve and respect several features during the denoising process. We demonstrate the strength of our method in terms of denoising, feature preservation and computational efficiency.

An iterative joint bilateral filtering for depth refinement of a 3D model

Time-of-Flight (TOF) sensor is an acquisition device that can capture the range data using an infrared pulse in real-time; however, it is unsuitable to perform fine 3D scanning due to sensing limitations such as low depth resolution and severe random noise. The depth measurements between adjacent regions are not distinct so that the reconstructed geometry often fails to express the details of an original shape. To perform noise removal without a loss of details, an edge-preserving filter by [Tomasi and Manduchi 1998]] has been adopted. The joint bilateral filter [J. Kopf 2007] is also used since it provides a better performance when additional information such as a high resolution image is available. However, the quality of acquired depth data needs more improvement to generate a more precise 3D model. In this research, an iterative algorithm is proposed based on the coarse-to-fine strategy to refine the depth information acquired from TOF sensor. Our method focuses on the reconstruction of 3D geometry from the measured depth map and color images. Our method outperforms other filtering techniques in noise reduction and the refinement of raw data.

Rendering of human skin during physical exercise

Many researchers have shown interest in the realistic rendering of a human face since it is crucial to the success of many applications in computer graphics, games and animation. However, it is difficult to represent the realistic appearance of a facial skin since it has complex physical properties and its appearance tends to vary as the situation changes. For example, the glossiness of a facial skin increases after exercise and the color of it becomes red after drink. In this paper, we introduce a method to simulate the appearance change of a human skin during physical exercise. Our key idea is that the appearance of a skin under exercise can be divided into a surface reflectance component and a subsurface scattering component so that they are approximated by a BRDF model and a BSSRDF model, respectively. The Torrance-Sparrow BRDF model describes the glossiness and the roughness that are affected by sweat and oil on the skin surface. The Multipole model [Donner and Jensen 2005] with absorption σa and scattering σs parameters represents subsurface scattering effects that occur between cells and pigments under the surface. In this paper, the parameters of each reflectance model are acquired by measurement.

Fast multiview three-dimensional reconstruction method using cost volume filtering

Abstract. As the number of customers who want to record three-dimensional (3-D) information using a mobile electronic device increases, it becomes more and more important to develop a method which quickly reconstructs a 3-D model from multiview images. A fast multiview-based 3-D reconstruction method is presented, which is suitable for the mobile environment by constructing a cost volume of the 3-D height field. This method consists of two steps: the construction of a reliable base surface and the recovery of shape details. In each step, the cost volume is constructed using photoconsistency and then it is filtered according to the multiscale. The multiscale-based cost volume filtering allows the 3-D reconstruction to maintain the overall shape and to preserve the shape details. We demonstrate the strength of the proposed method in terms of computation time, accuracy, and unconstrained acquisition environment.

Multi-scale feature matching between 2D image and 3D model

Although high performance scanners and design tools are rapidly developed, users still suffer from generating desired 3D contents. Furthermore, it requires advanced skills of design tools and considerable times to produce a satisfactory result. Recently, being motivated by these shortcomings of 3D content generation, various approaches have been studied to generate and manipulate a 3D content conveniently. One of them is using 2D input data such as sketches and real photos to handle the 3D templates. Since this approach is intuitive to the users, it can be efficiently used to manipulate existing 3D contents. Moreover, a large number of new 3D contents can be created by using billions of existing 2D images on online.

Deformation and Simulation of 3D Contents for the Digilog Book

Recently the use of e-book has been rapidly increased due to the development of portable electronic devices taking advantage of their portability and storage. The e-book can contain various multimedia contents which cannot be expressed in a traditional text-book such as movie clips, music and flash games. However, physical presence and analog sensibility of a traditional book cannot be described in the e-book. To overcome these limitations, the concept of digilog book has been proposed, which combines analog sensibility and digitized contents. In this research, we propose 3D contents deformation and simulation techniques in the Augmented Reality(AR) environments that can be utilized in the digilog book. Furthermore, the deformation technique is extended to dynamic simulation and user specified simulation using a physics engine and deformation paths.

Enhancement of range data using a structure-aware filter

Range data acquired by affordable IR depth sensors are significantly noisy and lack shape detail, making it difficult to obtain high-quality scans using depth cameras. Most methods to enhance noisy data do not preserve the original structure of the measured scene during data refinement, particularly at sharp edges and object boundaries. We present a novel approach to enhance the range data in a structure-aware manner based on the normal information in a scene. We first computed reliable normals by smoothing out noise and sharpening the edges of individual points using color-guided normal smoothing. Based on the normal information, the noisy range data were refined by a 3D bilateral filtering technique while preserving ridges, valleys, and depth discontinuities. In a comparison, our method outperformed previous techniques in terms of noise suppression and faithful structure reconstruction. Graphical abstractDisplay Omitted HighlightsWe achieve color-guided normal smoothing from a noisy depth map, which can successfully obtain a reliable normal field.We develop a structure-preserving range data enhancement algorithm based on normal information through a 3D bilateral filter.The proposed method can inherently circumvent common artifacts such as texture copy and edge blurring by avoiding using color information in depth adjustment.