Kyoung-Mi Lee

An adaptive resource-allocating network for automated detection, segmentation, and classification of breast cancer nuclei topic area: image processing and recognition

This paper presents a unified image analysis approach for automated detection, segmentation, and classification of breast cancer nuclei using a neural network, which learns to cluster shapes and to classify nuclei. The proposed neural network is incrementally grown by creating a new cluster whenever a previously unseen shape is presented. Each hidden node represents a cluster used as a template to provide faster and more accurate nuclei detection and segmentation. Online learning gives the system improved performance with continued use. The effectiveness of the resulting system is demonstrated on a task of cytological image analysis, with classification of individual nuclei used to diagnose the sample. This demonstrates the potential effectiveness of such a system on diagnostic tasks that require the classification of individual cells.

Component-based face detection and verification

For face detection, while pose and illumination significantly change the global facial appearance, components of a face are less affected by these changes. Component detectors can accurately locate facial components, and component-based approaches can be used to check whether the geometric locations of the components comply with a face. This paper proposes a face detection and verification method using component-based online learning, which is based on using unsupervised clustering to find a set of templates specific to faces consisting of face component and their relations. The main difference from previously reported component-based face detection methods is the use of online learning, which is ideal for highly repetitive tasks. This results in faster and more accurate face detection, because system performance improves with continued use. Further, uncertainty is added by calculating the standard deviation of face components and their relations. A component-based method with uncertainty provides flexibility to allow variability to describe an object in appearance and geometry.

Interactive Game Designed for Early Child using Multimedia Interface : Physical Activities

This paper proposes interactive game elements for children : contents, design, sound, gesture recognition, and speech recognition. Interactive games for early children must use the contents which reflect the educational needs and the design elements which are all bright, friendly, and simple to use. Also the games should consider the background music which is familiar with children and the narration which make easy to play the games. In gesture recognition and speech recognition, the interactive games must use gesture and voice data which hits to the age of the game user. Also, this paper introduces the development process for the interactive skipping game and applies the child-oriented contents, gestures, and voices to the game.

Neural Network-Generated Image Retrieval and Refinement

The motivation for this work is to develop an image retrieval system that can discriminate between images and that can learn user’s preference with feedback to make more intelligent. This paper proposes a neural network to extend prototype refinement which retains information fed by users. The proposed three-layered neural network indexes an image database and makes clusters by an unsupervised approach at a hidden layer. Given a query, the neural system retrieves similar images by computing similarities with images in the near clusters by a supervised approach at an output layer. To provide preference, users can select some images as relevant ones or irrelevant ones. With this feedback, the proposed refinement method estimates global approximations of radial-basis functions centered, and simultaneously adjusts corresponding prototypes. The system demonstrated the effectiveness of prototype refinement generated by the proposed neural network.

Silhouette-based human motion estimation for movement education of young children

To estimate a human motion, in this paper, we propose a neural approach using silhouettes in video frames captured by two cameras placed at the front and side of the human body. To extract features of the silhouettes for motion estimation, the proposed system computes both global and local features and then groups these features into static and dynamic features depending on whether features are in a static frame. Extracted features are used to train a RBF network. The neural system uses static features as the input of the neural network and dynamic features as additional features for classification. In this paper, the proposed method was applied to movement education for young children. The basic movements for such education consist of locomotor movements, such as walking, jumping, and hopping, and non-locomotor movements, including bending, stretching, balancing and turning. The system demonstrated the effectiveness of motion estimation for movement education generated by the proposed neural network.

A new EM algorithm for resource allocation network

Clustering usually assumes that the number of clusters is known or given. No knowledge of such a priori information is needed to find an appropriate number of clusters. This paper introduces an elliptical clustering algorithm with incremental growth of clusters, which is derived from the batch EM algorithm with a decay factor and a novelty criterion. The proposed algorithm can start with no or a small number of clusters. Whenever unusual data is presented, the algorithm adds a new cluster and finally the number of clusters in the data is obtained after clustering. The usefulness of the proposed algorithm is demonstrated for texture image segmentation and skin image segmentation.

Component-based human body tracking for posture estimation

To track a human body and estimate its posture, a component-based approach is less susceptible to changes in posture and lighting. This paper proposes model-based tracking with a component-based human body model comprised of 10 components and their hierarchical link. The proposed method first divides a video frame into blobs based on color, groups the blobs to make components, and matches the components to human body parts. Instead of matching blobs individually, the proposed model-based tacking uses components and their links. This paper shows the making of coarse-to-fine searches, so it offers a model that can make human-body matching more time-efficient.

Adaptive estimation of human posture using a component-based model

To detect a human body and recognize its posture, a component-based approach is less susceptible to changes in posture and lighting conditions. This paper proposes a component-based human-body model that comprises ten components and their flexible links. Each component contains geometrical information, appearance information, and information on the links with other components. The proposed method in this paper uses hierarchical links between components of human body, so that it allows to make coarse-to-fine searches and makes human-body matching more time-efficient. To adaptively estimate the posture in change of posture and illumination, we update the component online every time a new human body is incoming.

Intelligent tracking persons through non-overlapping cameras

An intelligent surveillance system can judge and handle a situation automatically within a wide monitoring area and unattended environment that has no certain human supervisor. In this paper, we propose a way to track persons through non-overlapping cameras that are connected over a network with a server. To track persons with a camera and send the tracking data to other cameras, the proposed system uses a human model that comprises a head, a torso, and legs. Also, with a trajectory model, the proposed system can predict the probability which an exited person from one camera is incoming to other cameras. The system is updated online during the lifetime of the system. These enable the proposed to keep tracking the recognized person in a wide area, to provide a guide for monitoring multiple cameras, and to adapt changes with time.

Component-based online learning for face detection and verification

Component detectors can accurately locate facial components, and component-based approaches can be used to build detectors that can handle partial occlusions. This paper proposes a face detection and verification method using component-based online learning. The main difference from previously reported component-based approaches is the use of online learning, which is ideal for highly repetitive tasks. This results in faster and more accurate face detection, because system performance improves with continued use. Further, uncertainty is added by calculating the standard deviation of face components and their relations.