Pedestrian detection is a key problem in night vision processing with a dozen of applications that will positively impact the performance of autonomous systems. Despite significant progress, our study shows that performance of state-of-the-art thermal image pedestrian detectors still has much room for improvement. The purpose of this paper is to overcome the challenge faced by the thermal image pedestrian detectors, which employ intensity based Region Of Interest (ROI) extraction followed by feature based validation. The most striking disadvantage faced by the first module, ROI extraction, is the failed detection of cloth insulted parts. To overcome this setback, this paper employs an algorithm and a principle of region growing pursuit tuned to the scale of the pedestrian. The statistics subtended by the pedestrian drastically vary with the scale and deviation from normality approach facilitates scale detection. Further, the paper offers an adaptive mathematical threshold to resolve the problem of subtracting the background while extracting cloth insulated parts as well. The inherent false positives of the ROI extraction module are limited by the choice of good features in pedestrian validation step. One such feature is curvelet feature, which has found its use extensively in optical images, but has as yet no reported results in thermal images. This has been used to arrive at a pedestrian detector with a reduced false positive rate. This work is the first venture made to scrutinize the utility of curvelet for characterizing pedestrians in thermal images. Attempt has also been made to improve the speed of curvelet transform computation. The classification task is realized through the use of the well known methodology of Support Vector Machines (SVMs). The proposed method is substantiated with qualified evaluation methodologies that permits us to carry out probing and informative comparisons across state-of-theart features, including deep learning methods, with six standard and in-house databases. With reference to deep learning, our algorithm exhibits comparable performance. More important is that it has significant lower requirements in terms of compute power and memory, thus making it more relevant for depolyment in resource constrained platforms with significant size, weight and power constraints. (C) 2016 Elsevier B.V. All rights reserved.

Content Based Retrieval (CBR) systems use Relevance Feedback (RF) to fill the semantic gap. RF can be short-term or long-term. The introduction of long-term learning methods address the memory problem in short-term learning methods. In this letter we propose a new method to enhance the gain of long-term relevance feedback. We have come up with a long term learning scheme in relevance feedback for CBR. The proposed system integrates the user feedback from all iterationations and instills memory into the feedback system of CBR without saving any log of earlier retrievals. In this letter, we have come up with a method to update the cluster parameters and weights assigned to features by accumulating the knowledge obtained from the user over iterations. The proposed update method is validated in the image retrieval context in terms of conventional recall-precision graph and retrieval accuracy.

In image retrieval, Curvelet global features have been used so far. They have shown promising results in characterizing texture because of its inherent ability to capture edge information more accurately than Wavelet and Gabor. Global feature fails to characterize the local features of the images. So, we have proposed a technique to combine the global texture (Curvelet) and color features with local features derived from Salient regions. We present a Salient region detector (based on Curvelet) that extracts the regions where variations occur. We show that using the global distribution of local features in addition to global features provides better retrieval performance than those features which represents the global nature of the image alone.

We present a new inhomogeneous image restoration model with an edge detection based regularization term, for known linear shift invariant blur kernel. Our regularization term penalizes the noise, but not the edges. Our method is well suited for real world images with texture. The proposed algorithm has two major benefits : It is tuned to decouple edges from noise and it is computationally very simple. Our method is validated qualitatively and quantitatively against state of the art restoration methods with extensive experiments.

An efficient non-orthogonal pyramid representation was proposed by Burt. However it has been stated in literature that the Laplacian sub bands of Burt pyramid have redundant information. In this paper, we propose a modified pyramid representation to reduce the redundancy in Laplacian sub bands. The proposed pyramid representation makes use of well studied de-blurring algorithm to get a prediction of blurred Gaussian images. The proposed pyramid is an improvement on Burt pyramid as it exhibits reduced sub-band frequency overlap, cross correlation and mutual information. The advantage of using this pyramid representation in image magnification and progressive image transmission (PIT) in noisy and unreliable networks is discussed here.

Accurate image restoration is of paramount importance for low-level vision, computer vision, and various other fields. Numerous complex restoration algorithms have been stated in the literature. Performance of these restoration algorithms differs with the nature of the image and distortion. These algorithms are evaluated either qualitatively or quantitatively by comparing the restored image with the original image. The practical drawback of this quantitative comparison is the requirement for the original image. In this paper, we propose a measure, with theoretical grounding, to objectively evaluate the restoration algorithms with no knowledge about the original image. This measure analyses the deblurring as well as the denoising nature of the restoration method. The main utility of this measure will be in designing automatic restoration systems. Effectiveness of this measure is substantiated with experimental results.