Huitao Luo

Optimization design of cascaded classifiers

We present two optimization algorithms for the design of a cascade of classifiers, which is becoming a popular choice for many classification problems. Both algorithms represent each node classifier of a cascade using a high-level abstraction model and attempt to jointly optimize the setting of the thresholding parameters of all the node classifiers within the cascade. We applied both algorithms to optimize the famous Viola and Jones face detector and one of them in particular greatly improved the performance. We believe both algorithms can serve as a useful post-processing process for general cascaded classifier design.

An efficient automatic redeye detection and correction algorithm

A fully automatic redeye detection and correction algorithm is presented to address the redeye artifacts in digital photos. The algorithm contains a redeye detection part and a correction part. The detection part is modeled as a feature based object detection problem. Adaboost is used to simultaneously select features and train the classifier. A new feature set is designed to address the orientation-dependency problem associated with the Haar-like features commonly used for object detection design. For each detected redeye, a correction algorithm is applied to do adaptive desaturation and darkening over the redeye region.

A training-based no-reference image quality assessment algorithm

We present a new image quality assessment algorithm that does not rely on reference images. Our general framework is to emulate human quality assessment by first detecting visual components, then assessing quality against an empirical model. We describe an instance of this framework where visual component detection is realized as a face detection method, and quality modeling is realized using radial basis function (RBF) networks. Experiments with this prototype system yielded promising results.

Optimal selection of segmentation algorithms based on performance evaluation

Image segmentation is a critical task and many algorithms have been proposed in the literature. However, how to select the best technique when facing a real application is still an open issue. A novel method for optimal selection of segmentation algorithms is introduced and an expert system for determining the best algorithms using evaluation results is proposed. The available segmentation algorithms are first evaluated and their characteristics or behaviors in a number of situations are then determined. The determination of the best algorithm is thus based on the knowledge obtained by objective evaluation of available segmentation algorithms. In other words, the algorithm selection is made according to the properties of the algorithms and images to be segmented. Using this idea, a prototype expert system for algorithm selection is designed and implemented. The design considerations and data structure of this system are detailed. Experiments with various real images are carried out and the results indicate that such an idea is feasible and such a system is suitable for practical segmentation applications.

On face detection in the compressed domain

We propose a fast face detection algorithm that works directly on the compressed DCT domain. Unlike the previous DCT domain processing designs that are mainly based on skin-color detection, our algorithm analyzes both color and texture information contained in the DCT parameters, therefore could generate more reliable detection results. Our texture analysis is mainly based on statistical model training and detection. A number of fundamental problems, e.g., block quantization, preprocessing in the DCT domain, and feature vector selection and classification in the DCT domain, are discussed.

Designing an interactive tool for video object segmentation and annotation

An interactive authoring system is proposed for semi-automatic video object segmentation and annotation. This system features a new contour interpolation algorithm, which enables the user to define the contour of a video object on multiple frames while the computer interpolates the missing contours of this object on every frame automatically. Typical active contour (snake) model is adapted and the contour interpolation problem is decomposed into two directional contour tracking problems and a merging problem. In addition, new user interaction models are created for the user to interact with the computer. Experiments indicate that this system offers a good balance between algorithm complexity and user interaction efficiency.

Model-based segmentation and tracking of head-and-shoulder video objects for real time multimedia services

A statistical model-based video segmentation algorithm is presented for head-and-shoulder type video. This algorithm uses domain knowledge by abstracting the head-and-shoulder object with a blob-based statistical region model and a shape model. The object segmentation problem is then converted into a model detection and tracking problem. At the system level, a hierarchical structure is designed and spatial and temporal filters are used to improve segmentation quality. This algorithm runs in real time over a QCIF size video, and segments it into background, head and shoulder three video objects on average Pentium PC platforms. Due to its real time feature, this algorithm is appropriate for real time multimedia services such as videophone and web chat. Simulation results are offered to compare MPEG-4 performance with H.263 on segmented video objects with respects to compression efficiency, bit rate adaptation and functionality.

Rubberband: an improved graph search algorithm for interactive object segmentation

We present a novel interactive object segmentation algorithm: Rubberband. Rubberband is motivated by graph search based boundary detection algorithms but has greatly improved robustness and controllability performance with (1) a novel user interface; (2) an improved graph search algorithm; (3) an automatic scale selection algorithm.

An interactive authoring system for video object segmentation and annotation

An interactive authoring system is proposed for semi-automatic video object (VO) segmentation and annotation. This system features a new contour interpolation algorithm, which enables the user to define the contour of a VO on multiple frames while the computer interpolates the missing contours of this object on every frame automatically. Typical active contour (snake) model is adapted and the contour interpolation problem is decomposed into a two-directional contour tracking problem and a merging problem. In addition, new user interaction models are designed for the user to interact with the computer. Experiments indicate that this system offers a good balance between algorithm complexity and user interaction efficiency.

Statistical model based video segmentation and its application to very low bit rate video coding

We present a statistical model-based video segmentation algorithm for typical videophone and video-conference applications. This algorithm makes use of online information and tries to build statistical models for both background and foreground and update the models on the fly. A hierarchical system structure is designed and segmentation is combined with tracking. Two possible applications are discussed: to generate VOP for MPEG-4 and to introduce subjective rate control for DCT-based algorithms. A rate distortion based rate control algorithm for H.263 is proposed and implemented as an example.