High dynamic range (HDR) images require tone-mapping to be viewed on low dynamic range (LDR) displays. The performance of tone-mapping algorithms can be evaluated through a subjective study in which participants based on their liking rank or score tone-mapped images (TMIs). Subjective evaluation can be painstakingly slow; therefore, several quantitative metrics have been proposed for objective evaluation. This article presents a new robust metric that uses 16 features, measuring the loss of color, contrast, brightness, and structure, extracted from the test TMI and the reference HDR image. The effect of these attributes on image quality is investigated and combined into a single score in the [0, 1] range describing the quality of TMI. We validate the performance of the proposed metric by comparing it with 24 existing state-of-the-art metrics. The study uses two subjective datasets of TMIs, including one existing benchmark dataset and a new proposed dataset comprising HDR images of a variety of scenes, and a dataset of traditional images not generated through tone-mapping. In these studies, our method shows the highest correlation with subjective scores for both datasets of TMIs and remains in the second position for the dataset of traditional images.

Tone mapping is necessary for Low Dynamic Range (LDR) devices to display High Dynamic Range (HDR) images and videos. Multiple video Tone Mapping Operators (vTMOs) have been devised for HDR videos. The majority of vTMOs apply an image TMO to each video frame. This is followed by pre/post filtering to ensure temporal coherence. However, this destroys the natural temporal variation in intensity that is inherent in a changing scene. Furthermore, in these methods computational complexity of an image TMO is scaled up in proportion to the number of frames. We propose an efficient method to extend an image TMO to video TMO. The proposed method is general and takes care of temporally coherent intensity variation between frames while addressing the well-known issue of flickering in tone mapped video. Additionally, it lowers computational complexity as a new tone mapping curve (TMC) is not generated on per frame basis. The proposed vTMO can be used to extend any state-of-the-art global TMO that is deemed to generate a TMC. Fresh TMC is generated only when a hard cut in video is detected or the global change in illumination in HDR video becomes large enough. Visual comparisons and objective evaluations with three well known image TMOs demonstrate that the suggested extension method generates high quality LDR video generally at low computational cost when compared to existing vTMOs. Further work on efficient implementation of embedding an image TMO in our vTMO algorithm is expected to yield even better computational efficiency.

Tone-Mapping Operators (TMOs) are usually used to produce Low Dynamic Range (LDR) versions of the High Dynamic Range (HDR) images that encapsulate better perceptual quality. Many TMOs have been proposed in literature as displaying HDR images on LDR screen is one major requirement, as long as LDR displays are in use. On the other hand, there are also many high performance current Computer Vision System (CVS) applications in use such as those for Optical Character Recognition (OCR), Automated Driver Assistance System (ADAS), etc. that are based on processing LDR images. These are incapable of processing the vast dynamic range of HDR imaging. Many times performance of these systems is excellent when images with optimal illumination are used and degrade badly for underexposed or overexposed images. HDR imaging do provide the avenue to combat non-optimal lighting condition and generate LDR images that maximizes image features and quality that is best suited for the targeted CVS application. In this work we have used a high performance state-of-the-art TMO (ATT) algorithm and has designed an automated training based mechanism to tune its parameters for optimizing the performance of a popular and free online available OCR engine Tesseract. Here we focus on a specific scenario of vision-based industrial product inspection. LDR imaging in a batch processing of product-line may suffer from glare, light source in background or poor illumination due to variation in position, orientation and movement of the product on conveyor belt based system with respect to camera and light source. HDR imaging can help reduce false rejection if optimal LDR images can be obtained for text reading by the OCR engine. Our proposed learning-based Application Specific TMO (ASTMO) significantly improves the OCR accuracy for diverse illumination conditions. Though we are reporting early results but they can pave the way for further research in this direction.

Subjective evaluation of tone-mapped images is tedious and time-consuming; therefore, it is desirable to have algorithms for automatic quality assessment. Many full-reference and blind metrics have been developed for this purpose, but their performance is generally evaluated on limited benchmark datasets. This leaves a possibility that the observed performance of the metric could be due to overfitting, and it might indeed not perform well for all scenes. In this work, we propose a novel framework using population-based metaheuristics to evaluate the performance of these metrics without requiring any subjectively evaluated reference dataset. The proposed algorithm does not modify the individual image pixels, instead, the tone-mapping curve is modified to synthesize realistic tone-mapped images for evaluation. Moreover, it is not required to know the underlying model of the evaluated metric, which is treated just like a black box and can be replaced by any other metric seamlessly. Therefore, any new metrics designed in the future can also be easily evaluated by simply replacing just one module in the proposed evaluation framework. We evaluate six existing metrics and synthesize images to which the metrics fail to assign appropriate scores for visual quality. We also propose a method to rank the relative performance of evaluated metrics, through a competition in which each metric tries to find the errors in the scores given by other metrics.

Automatic License Plate Recognition (ALPR) for years has remained a persistent topic of research due to numerous practicable applications, especially in the Intelligent Transportation system (ITS). Many currently available solutions are still not robust in various real-world circumstances and often impose constraints like fixed backgrounds and constant distance and camera angles. This paper presents an efficient multi-language repudiate ALPR system based on machine learning. Convolutional Neural Network (CNN) is trained and fine-tuned for the recognition stage to become more dynamic, plaint to diversification of backgrounds. For license plate (LP) detection, a newly released YOLOv5 object detecting framework is used. Data augmentation techniques such as gray scale and rotatation are also used to generate an augmented dataset for the training purpose. This proposed methodology achieved a recognition rate of 92.2%, producing better results than commercially available systems, PlateRecognizer (67%) and OpenALPR (77%). Our experiments validated that the proposed methodology can meet the pressing requirement of real-time analysis in Intelligent Transportation System (ITS).

Quality assessment of tone-mapped images has been an active research area over the past years. Humans have a personal liking of image attributes such as vividness of colors, mean brightness level, and contrast. Therefore, image quality is often determined through subjective studies. The participants are asked to assign scores to the images, and the average scores are used to determine the relative quality of the images or the algorithms that produced them. Several metrics have been proposed that try to replicate this process and assign quantitative scores to the images. While these metrics perform reasonably in general and their scores correlate well with subjective scores, they can fail badly in some situations. This paper proposes a novel method to put these metrics to test under extreme conditions and observe their performance. For this, we use the differential evolution approach that keeps modifying the tone-mapping function iteratively, such that the assigned score by the metric keeps increasing over iterations. We show that visible distortions start showing up in the image in many situations, yet the score given by the metric remains very high.

Virtual try-on of clothes has received much attention recently due to its commercial potential. It can be used for online shopping or intelligent recommendation to narrow down the selections to a few designs and sizes. In this paper, we present a mixed reality system for 3D virtual clothes try-on that enables a user to see herself wearing virtual clothes while looking at a mirror display, without taking off her actual clothes. The user can select various virtual clothes for trying-on. The system physically simulates the selected virtual clothes on the user's body in real-time and the user can see virtual clothes fitting on the her mirror image from various angles as she moves. The major contribution of this paper is that we automatically customize an invisible (or partially visible) avatar based on the user's body size and the skin color and use it for proper clothes fitting, alignment and clothes simulation in our virtual try-on system. We present three scenarios: i) virtual clothes on the avatar, ii) virtual clothes on the user's image and iii) virtual clothes on the avatar blended with the user's face image. We have conducted a user study to evaluate the effectiveness of these three solutions from the end user's perception of quality attributes, cognitive attributes and attitude towards using. The user study shows that among these three scenarios, the second one is most preferred by the users and for 50% of them the experience they had with our system was sufficient to make the purchase decision for the outfits they virtually tried-on.

This paper proposes a new method to automatically segment and re-color the clothing in an image sequence. Background and foreground are separated using background subtraction method with a Gaussian mixture model built for the static scene. Skin and face areas are detected and removed on the segmented foreground. A 2D histogram is constructed using the remaining pixels to model the probability distribution of the clothing chromaticity, which is then applied to each frame to find accurate clothing area using back projection method. The segmented clothing area is re-colored by mapping the hue value to a new one in the HSV color space to maintain the contrast of the clothing. Experiments on the real video captured with a monocular webcam are shown to demonstrate the effectiveness of the proposed algorithm for clothing segmentation and re-coloring.

A new scheme for back-projection of weights for mean shift based object tracking is proposed. Weights are calculated based on relative counts of histogram bins for each feature used in similarity assessment. A fusion scheme is proposed to combine the back-projected weights from different features, such that the dissimilarities between the object being tracked and the background are boosted. A mechanism is proposed to calculate the overall similarity between reference and candidate windows, without actually doing the back-projection, and just by few computations based on histogram bin counts. Moreover, the reference and candidate windows do not need to have the same size. The proposed scheme shows better tracking results compared to the traditional mean shift based tracking, especially in case of fast object movement and high background clutter.

Three-dimensional (3D) urban models, come with huge data size, mainly consisting of the details of geometry and texture of the objects, and simplification of both is often needed for efficient streaming, rendering and visualization. A large number of objects in 3D urban models are buildings and the image files representing their texture contain information of the walls, streets, doors, and windows etc., which have linear edges. We present a technique to identify the images representing the texture of these objects, and propose a compact format to encode them in much smaller size compared to the existing standard image formats.

Global tone mapping operators (TMOs) are better in preserving the naturalness and relative illumination of the HDR scene compared to the local TMOs. However, they cannot preserve the details like the local TMOs. We propose a human visual system (HVS) based histogram method for global tone mapping, which can better preserve the details than the existing global TMOs. Our method can be incorporated to all histogram based global methods. We compare our results with traditional histogram adjustment by Ward et al. [Larson97] and photographic TMO [Reinhard2002], and in most of the cases our method is better in preserving the details of the input HDR images.

With advent of high dynamic range (HDR) imaging techniques, it has been possible to capture natural scenes in larger details. HDR images are tone-mapped to lower dynamic range (LDR) versions for displaying on paper or a screen. Details lost during tone-mapping are important for certain existing and future applications, and need to be preserved. However, the size of HDR images is very large and that gives rise to need of effective encoding techniques. In this paper, we present an encoding scheme for HDR images, which significantly reduces their storage requirements, with a negligible loss of information. We model an HDR image as a piecewise linear function of its tone-mapped version. The tone-mapped image and the error in modeling are stored as LDR images, and these two along with the created model, approximate the HDR image. Comparison with the existing state of the art technique is given to show the effectiveness of our proposed scheme.

In this paper, we present a user identification technique based on face recognition for secure human-machine interaction. User's face is matched with the faces memorized by the machine, and if a match is found with a reliable matching score, the machine gets ready to accept the commands. Gabor Wavelet Transform coefficients are used as features for matching, and they are computed on a dedicated LSI to attain high computational speed. For matching, an algorithm based on the elastic graph matching is used, and that is also implemented on hardware. We also propose an arm tracking algorithm for communication with machines using arm gesture. The algorithm utilizes stereo vision to define search regions in left and right images iteratively, and looks for the arm posture by matching with a three dimensional four degrees-of-freedom kinematics-based arm model. Tracking procedure is computationally efficient, robust to small occlusions, and works in unconstrained environment, which makes it suitable for general applications in human-machine interaction.

In this paper, we propose the client-server system which accomplishes acquisition, transmission and reconstruction of the 3D shape at a remote location in real time. The server consists of stereo cameras and a PC cluster, in which each PC creates a range image using multiple color images obtained from the stereo cameras connected to it. All PCs on the server side work in synchronization and generate range images in parallel with each other, which are then transmitted to the client as a point cloud. On the client side, the 3D model is reconstructed using a received point cloud by the point-based rendering method. We propose the cylindrical mapping method, which removes overlaps and annoying artifacts efficiently before the rendering process. All the procedure of generation, transmission, and reconstruction of successive range images is done in real time.

Maximally flat (MAXFLAT) finite-impulse response digital filters, including the Hilbert transformers (HTs), have the smoothest magnitude responses among the available types of FIR digital filters. However, the transition bands of MAXFLAT filters are relatively wider, and this makes them unattractive in certain applications. We present new designs of even and odd length HTs by transforming a design of digital differentiators satisfying maximal linearity constraints at omega = pi/2. The resultant even and odd length HTs have highly smooth magnitude responses around omega = pi/2 and omega = (pi/4,3pi/4) respectively, and have relatively narrow transition bands compared to the existing MAXFLAT designs.

In this paper, we present a numerical technique for simplification of the texture of 3D maps. In the simplification process, we first select textures that can be segmented by straight lines using Hough transform. The selected texture images are simplified by the Hough transform. The other texture are filtered by the anisotropic diffusion and then the details are removed to reduce their data sizes. The representative colors of the image are chosen from the image, and the image is quantized based on those colors. The simplified image, with all major features of the original image preserved, has much smaller size as compared to the original.

MAXFLAT FIR digital filters has the smoothest and the most accurate magnitude responses among all the available types of FIR digital filters. However, the transition bands of MAXFLAT filters are relatively wider, and this makes them unattractive in certain applications. Traditionally, low/high pass MAXFLAT FIR filters are designed to satisfy MAXFLAT constraints at ends of the frequency band. In this paper, we present two new designs of halfband low/high pass filters that satisfy these constraints at the middle of the pass and stop bands. The first design is obtained by solving a system of linear equations obtained by applying MAXFLAT constraints to the magnitude response of the filter, while the second design is a transformation of one of our previous designs of maximally linear digital differentiators. Design examples show that the transition bands of the presented designs are narrower as compared to traditional designs

Three-dimensional (3D) urban models, come with huge data size and their simplification is often needed for efficient streaming, rendering and visualization. In this paper, we present a strategy for organization of this data such that the models of reduced resolution can be readily created. Simplification techniques for different types of objects in the urban models are presented and are used to represent them at different levels of detail. Objects are classified based on their importance relative to other objects in the model, and based on this classification, a suitable level of details of each object is determined for inclusion in the model of reduced resolution

Complete 3D surface models of real objects can be obtained by integrating several range images, each representing a different view of the object. The mesh generated for these models have irregular connectivity in general. Several remeshing techniques have been proposed to approximate the irregular mesh models with regular meshes, which have several advantages in several applications. We present a prototype of a system to generate regular 3D models of real objects with a simple setup of a single 3D scanner and a turntable. Multiple scans representing different parts of the object surface are mapped on a 2D plane, which facilitates the processes of smooth integration, hole filling and regularization. The complete 3D surface model is represented like a 2D image, on which existing and new signal and image processing techniques can be applied easily.