No abstract

We propose an algorithm to embed data directly in the bitstream of JPEG imagery. The motivation for this approach is that images are seldom available in uncompressed form. Algorithms that operate in spatial domain, or even in coefficient domain, require full (or at best) partial decompression. Our approach exploits the fact that only a fraction of JPEG code space is actually used by available encoders. Data embedding is performed by mapping a used variable length code (VLC) to an unused VLC. However, standard viewers unaware of the change will not properly display the image. We address this problem by a novel error concealment technique. Concealment works by remapping run/size values of marked VLCs so that standard viewers do not lose synchronization and displays the image with minimum loss of quality. It is possible for the embedded image to be visually identical to the original even though the two files are bitwise different. The algorithm is fast and transparent and embedding is reversible and file-size preserving. Under certain circumstances, file size may actually decrease despite carrying a payload.

Radar has established itself as an effective all-weather, day or night sensor. Radar signals can penetrate walls and provide information on moving targets. Recently, radar has been used as an effective biometric sensor for classification of gait. The return from a coherent radar system contains a frequency offset in the carrier frequency, known as the Doppler Effect. The movements of arms and legs give rise to micro Doppler which can be clearly detailed in the time-frequency domain using traditional or modern time-frequency signal representation. In this paper we propose a gait classifier based on subspace learning using principal components analysis(PCA). The training set consists of feature vectors defined as either time or frequency snapshots taken from the spectrogram of radar backscatter. We show that gait signature is captured effectively in feature vectors. Feature vectors are then used in training a minimum distance classifier based on Mahalanobis distance metric. Results show that gait classification with high accuracy and short observation window is achievable using the proposed classifier.

A through-the-wall radar image (TWRI) bears little resemblance to the equivalent optical image, making it difficult to interpret. To maximize the intelligence that may be obtained, it is desirable to automate the classification of targets in the image to support human operators. This paper presents a technique for classifying stationary targets based on the high-range resolution profile (HRRP) extracted from 3-D TWRIs. The dependence of the image on the target location is discussed using a system point spread function (PSF) approach. It is shown that the position dependence will cause a classifier to fail, unless the image to be classified is aligned to a classifier-training location. A target image alignment technique based on deconvolution of the image with the system PSF is proposed. Comparison of the aligned target images with measured images shows the alignment process introducing normalized mean squared error (NMSE) ≤ 9%. The HRRP extracted from aligned target images are classified using a naive Bayesian classifier supported by principal component analysis. The classifier is tested using a real TWRI of canonical targets behind a concrete wall and shown to obtain correct classification rates ≥ 97%.

No abstract