Soochahn Lee

Random tree walk toward instantaneous 3D human pose estimation

The availability of accurate depth cameras have made real-time human pose estimation possible; however, there are still demands for faster algorithms on low power processors. This paper introduces 1000 frames per second pose estimation method on a single core CPU. A large computation gain is achieved by random walk sub-sampling. Instead of training trees for pixel-wise classification, a regression tree is trained to estimate the probability distribution to the direction toward the particular joint, relative to the current position. At test time, the direction for the random walk is randomly chosen from a set of representative directions. The new position is found by a constant step toward the direction, and the distribution for next direction is found at the new position. The continual random walk through 3D space will eventually produce an expectation of step positions, which we estimate as the joint position. A regression tree is built separately for each joint. The number of random walk steps can be assigned for each joint so that the computation time is consistent regardless of the size of body segmentation. The experiments show that even with large computation gain, the accuracy is higher or comparable to the state-of-the-art pose estimation methods.

Extraction of Coronary Vessels in Fluoroscopic X-Ray Sequences Using Vessel Correspondence Optimization

We present a method to extract coronary vessels from fluoroscopic x-ray sequences. Given the vessel structure for the source frame, vessel correspondence candidates in the subsequent frame are generated by a novel hierarchical search scheme to overcome the aperture problem. Optimal correspondences are determined within a Markov random field optimization framework. Post-processing is performed to extract vessel branches newly visible due to the inflow of contrast agent. Quantitative and qualitative evaluation conducted on a dataset of 18 sequences demonstrate the effectiveness of the proposed method.

A new 3-D model retrieval system based on aspect-transition descriptor

In this paper, we propose a new 3-D model retrieval system using the Aspect-Transition Descriptor which is based on the aspect graph representation [1,2] approach. The proposed method differs from the conventional aspect graph representation in that we utilize transitions as well as aspects. The process of generating the Aspect-Transition Descriptor is as follows: First, uniformly sampled views of a 3-D model are separated into a stable and an unstable view sets according to the local variation of their 2-D shape. Next, adjacent stable views and unstable views are grouped into clusters and we select the characteristic aspects and transitions by finding the representative view from each cluster. The 2-D descriptors of the selected characteristic aspects and transitions are concatenated to form the 3-D descriptor. Matching the Aspect-Transition Descriptors is done using a modified Hausdorff distance. To evaluate the proposed 3-D descriptor, we have evaluated the retrieval performance on the Princeton benchmark database [3] and found that our method outperforms other retrieval techniques.

A Novel Cascade Classifier for Automatic Microcalcification Detection

In this paper, we present a novel cascaded classification framework for automatic detection of individual and clusters of microcalcifications (μC). Our framework comprises three classification stages: i) a random forest (RF) classifier for simple features capturing the second order local structure of individual μCs, where non-μC pixels in the target mammogram are efficiently eliminated; ii) a more complex discriminative restricted Boltzmann machine (DRBM) classifier for μC candidates determined in the RF stage, which automatically learns the detailed morphology of μC appearances for improved discriminative power; and iii) a detector to detect clusters of μCs from the individual μC detection results, using two different criteria. From the two-stage RF-DRBM classifier, we are able to distinguish μCs using explicitly computed features, as well as learn implicit features that are able to further discriminate between confusing cases. Experimental evaluation is conducted on the original Mammographic Image Analysis Society (MIAS) and mini-MIAS databases, as well as our own Seoul National University Bundang Hospital digital mammographic database. It is shown that the proposed method outperforms comparable methods in terms of receiver operating characteristic (ROC) and precision-recall curves for detection of individual μCs and free-response receiver operating characteristic (FROC) curve for detection of clustered μCs.

Forest Walk Methods for Localizing Body Joints from Single Depth Image.

We present multiple random forest methods for human pose estimation from single depth images that can operate in very high frame rate. We introduce four algorithms: random forest walk, greedy forest walk, random forest jumps, and greedy forest jumps. The proposed approaches can accurately infer the 3D positions of body joints without additional information such as temporal prior. A regression forest is trained to estimate the probability distribution to the direction or offset toward the particular joint, relative to the adjacent position. During pose estimation, the new position is chosen from a set of representative directions or offsets. The distribution for next position is found from traversing the regression tree from new position. The continual position sampling through 3D space will eventually produce an expectation of sample positions, which we estimate as the joint position. The experiments show that the accuracy is higher than current state-of-the-art pose estimation methods with additional advantage in computation time.

Selecting Representative Views of 3D Objects By Affinity Propagation for Retrieval and Classification

We propose a method to select representative views of single objects and classes of objects for 3D object retrieval and classification. Our method is based on projected 2D shapes, or views, of the 3D objects, where the representative views are selected by applying affinity propagation to cluster uniformly sampled views. Affinity propagation assigns prototypes to each cluster during the clustering process, thereby providing a natural criterion to select views. We recursively apply affinity propagation to the selected views of objects classified as single classes to obtain representative views of classes of objects. By enabling classification as well as retrieval, effective management of large scale databases for retrieval can be enhanced, since we can avoid exhaustive search over all objects by first classifying the object. We demonstrate the effectiveness of the proposed method for both retrieval and classification by experimental results based on the Princeton benchmark database [16].

Deep Learning Based Coronary Artery Motion Artifact Compensation Using Style-Transfer Synthesis in CT Images.

Motion artifact compensation of the coronary artery in computed tomography (CT) is required to quantify the risk of coronary artery disease more accurately. We present a novel method based on deep learning for motion artifact compensation in coronary CT angiography (CCTA). The ground-truth, i.e., coronary artery without motion, was synthesized using full-phase four-dimensional (4D) CT by applying style-transfer method because it is medically impossible to obtain in practice. The network for motion artifact compensation based on very deep convolutional neural network (CNN) is trained using the synthesized ground-truth. An observer study was performed for the evaluation of the proposed method. The motion artifacts were markedly reduced and boundaries of the coronary artery were much sharper than before applying the proposed method, with a strong inter-observer agreement (kappa = 0.78).

Nonrigid 2D registration of coronary artery angiograms with periodic displacement field

We propose a novel method for nonrigid registration of whole coronary artery sequence models with periodic displacement field. 2D nonrigid registration method is proposed that periodic deformed information is applied into correspondence with whole fluoroscopic angiograms. The proposed methodology is divided into two parts: one cycles nonrigid registration, spreading periodic displacement information into other cycles. In the first part, a nonrigid registration method of one cycle is implemented and used to compensate for any local shape discrepancy. In the second part, periodic displacement field is spreading into images on other cycles in order to align the whole sequence. Experimental evaluation conducted on a set of 9 fluoroscopic angiograms results in a reduced target registration error, which showed the effectiveness of the proposed methodology.

Nonrigid 2D registration of fluoroscopic coronary artery image sequence with propagated deformation field

We propose a novel method for nonrigid registration of coronary arteries within frames of a fluoroscopic X-ray angiogram sequence with propagated deformation field. The aim is to remove the motion of coronary arteries in order to simplify further registration of the 3D vessel structure obtained from computed tomography angiography, with the x-ray sequence. The Proposed methodology comprises two stages: propagated adjacent pairwise nonrigid registration, and, sequence-wise fixed frame nonrigid registration. In the first stage, a propagated nonrigid transformation reduces the disparity search range for each frame sequentially. In the second stage, nonrigid registration is applied for all frames with a fixed target frame, thus generating a motion-aligned sequence. Experimental evaluation conducted on a set of 7 fluoroscopic angiograms resulted in reduced target registration error, compared to previous methods, showing the effectiveness of the proposed methodology.

Consistent color and detail transfer from multiple source images for video and images

In this paper, we propose a method to jointly transfer the color and detail of multiple source images to a target video or image. Our method is based on a probabilistic segmentation scheme using Gaussian mixture model (GMM) to divide each source image as well as the target video frames or image into soft regions and determine the relevant source regions for each target region. For detail transfer, we first decompose each image as well as the target video frames or image into base and detail components. Then histogram matching is performed for detail components to transfer the detail of matching regions from source images to the target. We propose a unified framework to perform both color and detail transforms in an integrated manner. We also propose a method to maintain consistency for video targets, by enforcing consistent region segmentations for consecutive video frames using GMM-based parameter propagation and adaptive scene change detection. Experimental results demonstrate that our method automatically produces consistent color and detail transferred videos and images from a set of source images.