Quadratic Gabor filters for object detection

An Innovative Approach for Detection of Armoured Vehicle in Airborne Thermal Imagery Using Morphological Processing and Texture Feature Extraction

Automatic detection of a vehicle in an airborne thermal imagery is a challenging research topic in computer vision, especially the detection of military tanks in the field. Various methodologies for detection in forward-looking infrared imagery, which has higher spatial resolution, has been discussed by a number of researchers in literature. The algorithm we developed in the present study detects tanks not only in higher resolution but in lower resolution imagery as well. Detection algorithm is initiated by the segmentation of thermal image using mean shift, which provides possible targets present in the field other than the background. To reduce clutter and uneven illumination in a thermal image, a pre-processing morphological algorithm based on top-hat filtering has been implemented. After convolution of image window with Gabor filter banks, we extracted the energy feature of each image generated after convolution. The energy vector of such a target and the neighbouring background window has been calculated, and the similarity between the target and background using distance-measuring method has been measured. The minimum distance is used as the threshold to decide the target. A comparative study has been carried out between tanks and various targets/objects that appear similar to tanks in a thermal image. This validates our target detection algorithm. The false-positive rate and true-positive rate have been calculated for performance evaluation. Overall, this algorithm shows promising results for tank detection using single-band thermal imagery.

Personal object discovery in first-person videos

People know and care for personal objects, which can be different for individuals. Automatically discovering personal objects is thus of great practical importance. We, in this paper, pursue this task with wearable cameras based on the common sense that personal objects generally company us in various scenes. With this clue, we exploit a new object-scene distribution for robust detection. Two technical challenges involved in estimating this distribution, i.e., scene extraction and unsupervised object discovery, are tackled. For scene extraction, we learn the latent representation instead of simply selecting a few frames from the videos. In object discovery, we build an interaction model to select frame-level objects and use nonparametric Bayesian clustering. Experiments verify the usefulness of our approach.

Human identification using finger images

This paper presents a new approach to improve the performance of finger-vein identification systems presented in the literature. The proposed system simultaneously acquires the finger-vein and low-resolution fingerprint images and combines these two evidences using a novel score-level combination strategy. We examine the previously proposed finger-vein identification approaches and develop a new approach that illustrates it superiority over prior published efforts. The utility of low-resolution fingerprint images acquired from a webcam is examined to ascertain the matching performance from such images. We develop and investigate two new score-level combinations, i.e., holistic and nonlinear fusion, and comparatively evaluate them with more popular score-level fusion approaches to ascertain their effectiveness in the proposed system. The rigorous experimental results presented on the database of 6264 images from 156 subjects illustrate significant improvement in the performance, i.e., both from the authentication and recognition experiments.

Infrared target detection with probability density functions of wavelet transform subbands

We report the development of a wavelet multiresolution texture-based algorithm that uses the probability density functions (PDFs) of the subband of the wavelet decomposition of an image. The moments of these pdfs are used in a clustering algorithm to segment the targets from their background clutter. Using the tools of experimental methodology, we evaluate the performance of this algorithm on real infrared imagery under varying algorithm parameter sets as well as scene, image, and false-alarm conditions. We estimate a set of multidimensional predictive analytic performance models that relate the detection probabilities as functions of false alarm, algorithm internal parameter, target pixel number, target-to-background interference ratio, target-interference ratio, and Fechner-Weber and local entropy metrics in the scene. These models can be used to predict performance in regions were no data are available and to optimize performance by selection of the optimum parameter and constant false-alarm values in regions with known scene and metric conditions.