A new algorithm for estimating the multiple parameters (amplitude, phase and frequency) of a sinusoidal signal composed of the sum of multiple sinusoids and corrupted by additive white Gaussian noise is presented. This algorithm uses adaptive notch filters (ANFs) and a linear model. For the case where the frequencies of the sinusoids are among the unknown parameters, the ANFs first provide an accurate estimate of the frequency for each sinusoid. The linear model then uses the estimated frequencies to jointly estimate the corresponding amplitude and phase. For the case where the frequencies of the sinusoids are known but the amplitudes and phases are unknown, the linear model is used to obtain joint estimates of the amplitude and phase. Simulation examples are carried out to show the effectiveness of this algorithm. It is shown that the Cramer-Rao lower bound is asymptotically attained in most cases. The sensitivity of the linear model to the estimates of the ANF is also discussed.
This paper presents an object tracking system that uses mean shift with Interactive Multiple Model (IMM) Kalman filtering. The system represents targets using a weighted color histogram. The target window is masked with a Gaussian kernel to approximate its probability density function (PDF). The mean shift algorithm locates the mode of a PDF by ascending its gradient. The target is iteratively tracked in each video frame using the mean shift vector. Convergence occurs when the target and target candidate have similar color histograms. Each mean shift position estimate is used as a measurement in the IMM filter. The IMM algorithm runs two Kalman filters in parallel to estimate the target location in the next frame. The estimates from each filter are mixed to yield a weighted prediction. The motion models used by the IMM filter were constant velocity and constant acceleration. The results show that pairing the IMM and mean shift methods improves tracking accuracy and computation time.
In this paper, three different facial recognition algorithms are evaluated and compared, namely, principle component analysis (PCA), two-dimensional PCA (2D-PCA), and discrete cosine transform (DCT). The effect of the presence of the Gaussian and salt and Pepper noises was also considered during the evaluation of these algorithms. The results show that the best performance was obtained using the DCT algorithm with 92% dominant eigenvalues and 95.25 % accuracy.
Image forgery detection approaches are varied and serve same objectives. However, the difference in image properties causes some limitations of most of these approaches. Integrate multiple forensic approaches to increase the efficiency of detecting and localize the forgery was proposed based on the same image input source. In this paper, we propose a new detector algorithm based on different image source format. We propose approach to detect a copy-move forgery based on PatchMatch enhanced by the dense field technique. The F-measure score used same evaluation function to make the system more robust. The output result shows high efficiency of detecting and localizing the forgery in different image formats, for passive forgery detection.
