Ronda Venkateswarlu

Max-mean and max-median filters for detection of small targets

This paper deals with the problem of detection and tracking of low observable small-targets from a sequence of IR images against structural background and non-stationary clutter. There are many algorithms reported in the open literature for detection and tracking of targets of significant size in the image plane with good results. However, the difficulties of detecting small-targets arise from the fact that they are not easily discernable from clutter. The focus of research in this area is to reduce the false alarm rate to an acceptable level. Triple Temporal Filter reported by Jerry Silverman et. al., is one of the promising algorithms in this are. In this paper, we investigate the usefulness of Max-Mean and Max-Median filters in preserving the edges of clouds and structural backgrounds, which helps in detecting small-targets. Subsequently, anti-mean and anti-median operations result in good performance of detecting targets against moving clutter. The raw image is first filtered by max-mean/max-median filter. Then the filtered output is subtracted from the original image to enhance the potential targets. A thresholding step is incorporated in order to limit the number of potential target pixels. The threshold is obtained by using the statistics of the image. Finally, the thresholded images are accumulated so that the moving target forms a continuous trajectory and can be detected by using the post-processing algorithm. It is assumed that most of the targets occupy a couple of pixels. Head-on moving and maneuvering targets are not considered. These filters have ben tested successfully with the available database and the result are presented.

Estimating the eye gaze from one eye

Eye gaze estimation via images has been well investigated and all methods worked on images with the full face. This makes the eye details small and accuracy is affected. We address this problem by zooming in on a single eye. In this paper, we will show the validity of the method and investigate the performance with controlled synthetic data and also with real images. The principle is to rely on the fact that the outer boundary of the iris is a circle. With a fully calibrated camera, its elliptical image can be back-projected into the 3D space yielding two possible circles. To disambiguate, the solution is found by making use of anthropomorphic knowledge of the structure of the eyeball. Hence, getting a larger eye image with a zoom camera enabled us to achieve higher resolutions and thereby higher accuracies. The robustness of the algorithm was verified by extensive statistical trials conducted on synthetic data and real images. The two key contributions in this paper are to show that it is possible to estimate eye gaze with only one eye image and that consequently this achieves higher accuracy of eye gaze estimation.

Registration of infrared and visible-spectrum imagery for face recognition

A method is proposed to register visible and infrared face images. The vanishing-point based approach [J.G. Wang and E. Sung, 2001] is applied to the visible image to determine the 3D pose of the human head. Then the corresponding pose with respect to the infrared camera is computed through the known relationship by calibration between the visible and infrared cameras. By doing so, the skin temperature range within the infrared image can be superimposed over the visible face image. We use the EM strategy to first compute the 3D pose using some initially learned (PCA) model parameters, and then update iteratively the parameters for individual persons and their facial expressions till convergence. The EM technique models data uncertainty using Gaussian mixtures defined over positions and orientation of facial plane. The resulting weighted parameters estimation problem is solved using the Levenberg-Marquardt method. The results on the synthetic data and real images have verified the performance of the approach.

Adaptive mean and variance filter for detection of dim point-like targets

The problem of detecting small target in IR imagery has attracted much research effort over the past few decades. As opposed to early detection algorithms which detect targets spatially in each image and then apply tracking algorithm, more recent approaches have used multiple frames to incorporate temporal as well as spatial information. They often referred to as track before detect algorithms. This approach has shown promising results particularly for detection of dim point-like targets. However, the computationally complexity has prohibited practical usage for such algorithms. This paper presents an adaptive, recursive and computation efficient detection method. This detection algorithm updates parameters and detects occurrence of targets as new frame arrived without storing previous frames, thus achieved recursiveness. Besides, the target temporal intensity change is modeled by two Gaussian distribution with different mean and variance. The derivation of this generalized model has taken account of the wide variation of target speed, therefore detects wider range of targets.

Real-time Stereo Face Recognition by Fusing Appearance and Depth Fisherfaces

Reported 3D face recognition techniques assume the use of active 3D measurement for 3D facial capture. However, active method employ structured illumination (structure projection, phase shift, gray-code demodulation, etc) or laser scanning, which is not desirable in many applications. A major problem of using passive stereo is its lower 3D face resolution and thus no passive method for 3D face recognition has been reported. In this paper, a real-time passive stereo face recognition system is presented. Entire face detection, tracking, pose estimation and face recognition are investigated. We used SRI Stereo engine that outputs sub-pixel disparity automatically. An investigation is carried out in combining 3D and 2D information for face recognition. The straightforward two-stage principal component analysis plus linear discriminant analysis is carried out in appearance and depth face images respectively. A probe face is identified using sum of the weighted appearance and depth linear discriminant distances. We investigate the complete range of linear combinations to reveal the interplay between these two paradigms. The improvement of the face recognition rate using this combination is verified. The recognition rate by the combination is higher than that of either appearance alone or depth alone. We then discuss the implementation of the algorithm on a stereo vision system. A hybrid face and facial features detection/tracking approach is proposed which collects near-frontal views for face recognition. Our face detection/tracking approach automatically initializes without user intervention and can be re-initialized automatically if the tracking of the 3D face pose is lost. The experiments include two parts. Firstly, the performance of the proposed algorithm is verified on XM2VTS database; Secondly, the algorithm is demonstrated on a real-time stereo vision system. It is able to detect, track and recognize a person while walking toward a stereo camera.

Fusion of appearance and depth information for face recognition

In this paper, an investigation is carried out regarding combination of 3D and 2D information for face recognition. A two-stage method, PCA and a Reduced Multivariate Polynomial Model (RMPM), is developed to fuse the appearance and depth information of face images in feature level where simplicity (number of polynomial coefficients increases linearly with model-order and input-dimension, i.e. no dimension explosion as in the case of full multivariate polynomials)and ease of use (can be easily formulated into recursive learning fashion) are major concerns. To cater for fast on-line registration capability when a new user arrives, the learning is formulated into recursive form. The improvement of the face recognition rate using this combination is quantified. The recognition rate by the combination is better than either appearance alone or depth alone. The performance of the algorithm is verified on both XM2VTS database and a real-time stereo vision system, showing that it is able to detect, track and recognize a person walking towards a stereo camera within reasonable time.

EM Enhancement of 3D Head Pose Estimated by Perspective Invariance

In this paper, a new approach is proposed for estimating 3D head pose from a monocular image. The approach assumes the more difficult full perspective projection camera model as against most previous approaches that approximate the non-linear perspective projection via linear affine assumption. Perspective-invariance is used to estimate the head pose from a face image. Our approach employs a general prior knowledge of face structure and the corresponding geometrical constraints provided by the location of a certain vanishing point to determine the pose of human faces. To achieve this, eye-lines, formed from the far and near eye corners, and mouth-line of the mouth corners are assumed parallel in 3D space. Then the vanishing point of these parallel lines found by the intersection of the eye-line and mouth-line in the image can be used to infer the 3D orientation and location of the human face. Perspective invariance of cross ratio and harmonic range are used to locate the vanishing point stably. In order to deal with the variance of the facial model parameters, e.g. ratio between the eye-line and the mouth line, an EM framework is applied to update the parameters iteratively. We use the EM strategy to first compute the 3D pose using some initially learned (PCA) parameters, e.g. ratio and length, then update iteratively the parameters for individual persons and their facial expressions till convergence. The EM technique models data uncertainty as Gaussian defined over positions and orientation of facial plane. The resulting weighted parameters estimation problem is solved using the Levenberg-Marquardt method. The robustness analysis of the algorithm with synthetic data and real face images are included.

Multimode algorithm for detection and tracking of point targets

This paper deals with the problem of detection and tracking of point-targets from a sequence of IR images against slowly moving clouds as well as structural background. Many algorithms are reported in the literature for tracking sizeable targets with good result. However, the difficulties in tracking point-targets arise from the fact that they are not easily discernible from point like clutter. Though the point-targets are moving, it is very difficult to detect and track them with reduced false alarm rates, because of the non-stationary of the IR clutter, changing target statistics and sensor motion. The focus of research in this area is to reduce false alarm rate to an acceptable level. In certain situations not detecting a true target is acceptable, but declaring a false target as a true one may not be acceptable. Although, there are many approaches to tackle this problem, no single method works well in all the situations. In this paper, we present a multi-mode algorithm involving scene stabilization using image registration, 2D spatial filtering based on continuous wavelet transform, adaptive threshold, accumulation of the threshold frames and processing of the accumulated frame to get the final target trajectories. It is assumed that most of the targets occupy a couple of pixels. Head-on moving and maneuvering targets are not considered. It has been tested successfully with the available database and the results are presented.

Scale-space Based Corner Detection of Gray Level Images Using Plessey Operator

This paper proposes a multiscale corner detection method for gray level images based on scale-space theory and Plessey operator. The proposed method solves three problems existing in the original Plessey detector. First, it works in the scale-space domain, so it detects corners belonging to different scales instead of a certain scale. Second, only one parameter needs to be set instead of three parameters needed in the original Plessey method. Third, delocalization is a well-known inherent drawback of the Plessey corner operator and it will increase with the scale at which it operates. The proposed algorithm solves the problem by detecting the corners from small scale to large scale, then track back from large scale to small scale. As the delocalization in the smallest scale can be ignored, the proposed method obtain the accurate localization. This proposed multiscale scheme can also be applied to other spatial corner detectors to improve their performances. The simulation results and the application in stereo matching show the improved performance of the proposed method compared with the original Plessey detector and the SUSAN detector

Face tracking and recognition from stereo sequence

In this paper, we present a face recognition system that is able to detect, track and recognize a person walking toward a stereo camera. Integrating stereo and intensity pairs, pose can be detected and tracked. Face images, which are suitable for recognition, from stereo sequence can be selected automatically based on the pose. Then the straightforward Fisherface is carried out on the selected face images. Results of the tracking and recognition depict that this video based stereo approach provides a more robust performance since the 3D information is considered.