International Journal of Image and Graphics

Curve and surface-fitting are classic problems of approximation that find use in many fields, including computer vision. There are two broad approaches to the problem — interpolation, which seeks to fit points exactly, and regression, which seeks a rougher approximation which is more robust to noise. This survey looks at several techniques of both kinds, with a particular focus on applications in computer vision. We make use of an empirical first-level evaluation approach which scores the techniques on multiple features based on how important they are to users of the technique and developers. This provides a quick summary of the broad applicability of the technique to most situations, rather than a deep evaluation of the performance and accuracy of the technique obtained by running it on several datasets.

Computer vision-based recognition of Bangle vehicle license plates (LPs) is an arduous task in dirty and muddy situations. This paper proposes an efficient method for real-time computer vision-based recognition of Bangla vehicle LPs using contour analysis and prediction algorithms. The method initially applies gray scaling the input RGB images, histogram equalization to improve the grayscale image quality, edge detection using Sobel edge detector, and adaptive thresholding to convert it to a binary image. The system localizes the vehicle LP based on the maximum rectangular contour area and converts it into a predefined size. Noise removal technique using morphological dilation and erosion operation is used, followed by Gaussian filtering on binary image to improve the image quality further. The system clusters the two-lined LP into seven clusters. The sub-clustering is applied on specific clusters and makes 68 individual sub-clusters. The system extracts vector contour (VC) from each 68 individual classes. After VC extraction, the system normalizes it into a q predefined length. The system applies inter co-relation function (ICF) to categorize each sub-cluster to its previously defined individual class. For that, it calculates the maximum similarity between test and previously trained VCs. The system applies the dependency prediction algorithm in parallel to predict the whole string (district name) in the cluster-1 based on previously categorized class or classes (starting character or characters of the district part). (Metro) or (null) from cluster-2, “-” (hyphen) from cluster-3 and 6 are predicted automatically as their positions are fixed. The system is trained using 68 classes in which each class contains 100 samples generated by the augmentation technique. The system is tested using another set of 68 classes with a total of $68\times 100=6800$ images acquiring the recognition accuracy of 96.62% with the mean computational cost of 8.363ms/f. The system is also tested using 500 vehicle whole Bangla LPs achieving the mean whole LP recognition accuracy of 95.41% with a mean computational cost of 35.803ms/f.

A long range observing systems can be sturdily affected by scintillations. These scintillations are caused by changes in atmospheric conditions. In recent years, various turbulence mitigation approaches for turbulence mitigation have been exhibiting a promising nature. In this paper, we propose an effectual method to alleviate the effects of atmospheric distortion on observed images and video sequences. These sequences are mainly affected through floating air turbulence which can severely degrade the image quality. The existing algorithms primarily focus on the removal of turbulence and provides a solution only for static scenes, where there is no moving entity (real motion). As in the traditional SGL algorithm, the updated frame is iteratively used to correct the turbulence. This approach reduces the turbulence effect. However, it imposes some artifacts on the real motion that blurs the object. The proposed method is an alteration of the existing Sobolev Gradient and Laplacian (SGL) algorithm to eliminate turbulence. It eliminates the ghost artifact formed on moving object in the existing approach. The proposed method alleviates turbulence without harming the moving objects in the scene. The method is demonstrated on significantly distorted sequences provided by OTIS and compared with the SGL technique. The information conveyed in the scene becomes clearly visible through the method on exclusion of turbulence. The proposed approach is evaluated using standard performance measures such as MSE, PSNR and SSIM. The evaluation results depict that the proposed method outperforms the existing state-of-the-art approaches for all three standard performance measures.

Security has always been of paramount importance to humans. In the absence of a sense of security at one’s workplace, home or anywhere else, people feel uneasy and vulnerable. With the improvement of modern technology, along with the lack of time at hand, the need for faster, efficient, accurate as well as low-cost security techniques is more than ever. Image Captioning for Video Surveillance System is proposed to develop visual systems that generate contextual descriptions about objects in images, and then use these descriptions to provide information of the scene that needs to be secured. The proposed system uses a neural network model composed of a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) to caption the incoming video feed. The main significance of this paper is in integrating the system with Discrete Wavelet Transform (DWT), which is applied on the incoming video feed, so that the compressed LL band frames transferred wirelessly to the model are smaller in comparison, leading to less transfer time and faster processing by the model.

In recent times, there has been a tremendous progress in image dehazing for computer vision applications, while the sky region processed by these algorithms tends to degrade by noise and color distortion. In this paper, an improved dark channel prior algorithm is proposed which detects the sky region first and divides the image into sky region and non-sky region and then estimates the transmission of two parts separately, followed by combining with refining step. The proposed algorithm also accurately corrects the transmission of the sky region to avoid noise and color distortion. Experimental results show a greater quality improvement in the output images than the existing strategies.

In diverse computer applications, the analysis of image content plays a key role. This image content might be either textual (like text appearing in the images) or visual (like shape, color, texture). These two image contents consist of image’s basic features and therefore turn out to be as the major advantage for any of the implementation. Many of the art models are based on the visual search or annotated text for Content-Based Image Retrieval (CBIR) models. There is more demand toward multitasking, a new method needs to be introduced with the combination of both textual and visual features. This paper plans to develop the intelligent CBIR system for the collection of different benchmark texture datasets. Here, a new descriptor named Information Oriented Angle-based Local Tri-directional Weber Patterns (IOA-LTriWPs) is adopted. The pattern is operated not only based on tri-direction and eight neighborhood pixels but also based on four angles $0^{\circ }$, $45^{\circ }$, $90^{\circ }$, and $135^{\circ }$. Once the patterns concerning tri-direction, eight neighborhood pixels, and four angles are taken, the best patterns are selected based on maximum mutual information. Moreover, the histogram computation of the patterns provides the final feature vector, from which the new weighted feature extraction is performed. As a new contribution, the novel weight function is optimized by the Improved MVO on random basis (IMVO-RB), in such a way that the precision and recall of the retrieved image is high. Further, the proposed model has used the logarithmic similarity called Mean Square Logarithmic Error (MSLE) between the features of the query image and trained images for retrieving the concerned images. The analyses on diverse texture image datasets have validated the accuracy and efficiency of the developed pattern over existing.

Line drawing and screentoning are two distinct areas of study in non-photorealistic rendering, where the former emphasizes object contours, while the latter conveys tone and shading information on object surfaces. As these two problems are concerned with different yet equally important features, either method seldom delivers a complete description of the scene when used alone. Yet, research community has largely treated them as separate problems and thus resulted in two entirely different sets of solutions, complicating both implementation and usage. In this paper, we present a stylistic image binarization method called hybrid difference of Gaussians (HDoG) that performs both line drawing and screentoning in a unified framework. Our method is based upon two different extensions of DoG operator: one for line extraction, and the other for tone description. In particular, we propose an extension called adaptive DoG, that uses luminance as weight to automatically generate screentone that adapts to the local tone. Experimental results demonstrate that our hybrid method effectively generates aesthetically pleasing image binarizations that encompass both line drawing and screentoning, closely resembling professional pen-and-ink illustrations. Also, being based on Gaussian filtering, our method is very fast and also easy to implement.

HDR images are inherently very large in size compared to normal images. Hence, storage and communication overheads of HDR images are expensive to be used in mobile devices. Hence, invariably image compression is adopted for HDR images. In this paper, HDR image compression is achieved by down sampling the intensity levels while maintaining the dynamic range same as that of the original. This aspect retains the edge information of the images almost intact. Spatial down-sampling process is used to reduce the number of intensity samples. Consequently, this operation lowers the bit depth required to store the corresponding index file which in turn results in image compression.

Thresholding of Curvelet Coefficients, for image denoising, drains out subtle signal component in noise subspace. In effect, it also produces ringing artifacts near edges. We found that the noise sensitivity of Curvelet phases — in contrast to their magnitude — reduces with higher noise level. Thus, we preserved the phase of the coefficients below threshold at coarser scale and estimated the corresponding magnitude by Joint Bilateral Filtering (JBF) technique. In contrast to the traditional hard thresholding, the coefficients in the finest scale is estimated using Bilateral Filtering (BF). The proposed filtering approach in the finest scale exhibits better connectedness among the edges, while removing the granular artifacts in the denoised image due to hard thresholding. Finally, the use of Guided Image Filter (GIF) on the Curvelet-based reconstructed image (initial denoised image in spatial domain) ensures the preservation of small image information with sharper edges and textures detail in the final denoised image. The lower noise sensitivity of Curvelet phase at higher noise strength accelerates the performance of proposed method over several state-of-the-art techniques and provides comparable outcome at lower noise levels.

Documents are an essential source of valuable information and knowledge, and photographs are a great way of reminiscing old memories and past events. However, it becomes difficult to preserve the quality of such ancient documents and old photographs for an extremely long time, as these images usually get damaged or creased due to various extrinsic effects. Utilizing image editing software like Photoshop to manually reconstruct such old photographs and documents is a strenuous and an enduring process. This paper attempts to leverage the generative modeling capabilities of Conditional Generative Adversarial Networks by utilizing specialized architectures for the Generator and the Discriminator. The proposed Reminiscent Net has a U-Net-based Generator with numerous feature maps for complete information transfer with the incorporation of location and contextual details, and the absence of dense layers allows utilization of diverse sized images. Implementation of the PatchGAN-based Discriminator that penalizes the image at the scale of patches has been proposed. NADAM optimizer has been implemented to enable faster and better convergence of the loss function. The proposed method produces visually appealing de-creased images, and experiments indicate that the architecture performs better than various novel approaches, both qualitatively and quantitatively.

For handling digital images for various applications, image denoising is considered as a fundamental pre-processing step. Diverse image denoising algorithms have been introduced in the past few decades. The main intent of this proposal is to develop an effective image denoising model on the basis of internal and external patches. This model adopts Non-local means (NLM) for performing the denoising, which uses redundant information of the image in pixel or spatial domain to reduce the noise. While performing the image denoising using NLM, “denoising an image patch using the other noisy patches within the noisy image is done for internal denoising and denoising a patch using the external clean natural patches is done for external denoising”. Here, the selection of optimal block from the entire datasets including internal noisy images and external clean natural images is decided by a new hybrid optimization algorithm. The two renowned optimization algorithms Chicken Swarm Optimization (CSO), and Dragon Fly Algorithm (DA) are merged, and the new hybrid algorithm Rooster-based Levy Updated DA (RLU-DA) is adopted. The experimental results in terms of some relevant performance measures show the promising results of the proposed model with remarkable stability and high accuracy.

Image segmentation is a prime operation to understand the content of images. Multilevel thresholding is applied in image segmentation because of its speed and accuracy. In this paper, a novel multilevel thresholding algorithm based on Cuckoo search (CS) is introduced. One of the major drawbacks of metaheuristic algorithms is the stagnation phenomenon which leads to a fall into local optimums and premature convergence. To overcome this shortcoming, the idea of Darwinian theory is incorporated with CS algorithm to increase the diversity and quality of the individuals without decreasing the convergence speed of CS algorithm. A policy of encouragement and punishment is considered to lead searching agents in the search space and reduce the computational time. The algorithm is implemented based on dividing the population into specified groups and each group tries to find a better location. Ten test images are selected to verify the ability of our algorithm using the famous energy curve method. Two popular entropies criteria, Otsu and Kapur, are employed to evaluate the capability of the introduced algorithm. Eight different search algorithms are also implemented and compared with our method. Experimental results manifest that DCS is a powerful tool for multilevel thresholding and the obtained results outperform the CS algorithm and other heuristic search methods.

Due to the haze weather, the outdoor image quality is degraded, which reduces the image contrast, thereby reducing the efficiency of computer vision systems such as target recognition. There are two aspects of the traditional algorithm based on the principle of dark channel to be improved. First, the restored images obviously contain color distortion in the sky region. Second, the white regions in the scene easily affect the atmospheric light estimated. To solve the above problems, this paper proposes a single-image dehazing and image segmentation method via dark channel prior (DCP) and adaptive threshold. The sky region of hazing image is relatively bright, so sky region does not meet the DCP. The sky part is separated by the adaptive threshold, then the scenery and the sky area are dehazed, respectively. In order to avoid the interference caused by white objects to the estimation of atmospheric light, we estimate the value of atmospheric light using the separated area of the sky. The algorithm in this paper makes up for the shortcoming that the algorithm based on the DCP cannot effectively process the hazing image with sky region, avoiding the effect of white objects on estimating atmospheric light. Experimental results show the feasibility and effectiveness of the improved algorithm.

In digital images, Copy-Move Forgery is a general kind of forgery techniques. The process of replicating one part of the image within the same image is termed as copy-move forgery. An effective and reliable approach needs to be developed for identifying these forgeries for restoring the image trustworthiness. The main intent of this paper is to sort out the diverse copy-move image forgery detection models. This survey makes an effective literature analysis on a set of literal works from the past 10 years. The analysis is focused on categorizing the models based on transformation models, machine learning algorithms, and other advanced techniques. The main contribution and limitations of the works are clearly pointed out. In addition, the types of datasets and the simulation platforms utilized by different copy-move forgery detection (CMFD) models are analyzed. The performance measures evaluated by different contributions have been observed for making a concluding decision. The utilization of optimization algorithms on copy-move image forgery detection has also been studied. Finally, the research gaps and challenges with future direction are discussed, which is helpful for researchers in developing an efficient CMFD that could attain high performance.

Content-based image retrieval (CBIR) is a technique for images retrieval based on their visual features, i.e. induced by their pixels. The images are, classically, described by the image feature vectors. Those vectors reflect the texture, color or a combination of them. The accuracy of the CBIR system is highly influenced by the (i) definition of the image feature vector describing the image, (ii) indexing and (iii) retrieval process. In this paper, we propose a new CBIR system entitled ISE (Image Search Engine). Our ISE system defines the optimum combination of color and texture features as an image feature vector, including the Particle Swarm Optimization (PSO) algorithm and employing an Interactive Genetic Approach (GA) for the indexing process. The performance analysis shows that our suggested PCM (Proposed Combination Method) upgrades the average precision metric from 66.6% to 89.30% for the “Food” category color histogram, from 77.7% to 100% concerning CCVs (Color Coherence Vectors) for the “Flower” category and from 58% to 87.65% regarding the DCD (Dominant Color Descriptor) for the “Building” category using the Corel dataset. Besides, our ISE system showcases an average precision of 98.23%, which is significantly higher than other CBIR systems presented in related works.