Search

Handwritten recognition model characters have been reported by various researchers in past years and it becomes the most challenging day by day in this age of the digital world. Many machine learning algorithms have a great influence over the handwritten recognition system. In this segment, deep learning is one of the enhanced techniques to solve pattern recognition problems. Such complex problem we have imposed with convolutional neural network (CNN) for the handwritten characters. In this paper, we have made an attempt to build a recognition system for multi-scripts such as Odia and Bangla scripts. Here we have defined the data-driven learning mechanism of CNN along with deriving the discriminate features of handwritten images. This light-weighted CNN provides a feasible solution and reports a high recognition rate. Such deep learning-based approach is a new state of art method for developing an automatic recognition model and to meet real-time challenges.

Breast cancer is a leading cause of death after skin cancer and is a threat to women worldwide. The automated technique of breast cancer that is still being considered is screening X-ray mammography by radiologists, which sometimes led to false-positive and false-negative results, in turn increasing the mortality ratio in breast cancer patients. However, earlier study shows that AI can be used as a detecting aid for radiologists where either radiologists or AI can work hand in hand or AI can be used as the second opinion for better results. The main objective of this paper is to include a technical review, which mainly summarizes contemporary research progress on mammogram image analysis and an overview of various projected AI-based models such as machine learning (ML), deep learning (DL), transfer learning (TL) and extreme learning machine (ELM) and the way they function and their advantages and limitations. Also, we convey how the different models have taken into contemplation to detect breast cancer more accurately with lesser false-positive (sensitivity) or false-negative rates (specificity) using the 2D or 3D mammograms images. This article also discusses several hyperparameters that affect the model accuracy.

Transportation system (vehicle communication) plays a major role in today’s scenario. Detection of vehicle number plate exactly in blurry conditions was the most challenging issue found in the last three decades. Although many intensive studies were undertaken, none addressed this problem exhaustively. Various methods are introduced by several researchers for detecting the vehicle number from the vehicle number plate images. The purpose of this study was to investigate this current issue by implementing an edge-based approach on the basis of quantitative combination of Canny, Morphological and Sobel methods for the accurate detection of vehicle number in blurry conditions. The experimental results demonstrated that the proposed scheme outperforms its counterparts in terms of Sobel, Prewitt, Roberts, Laplacian of Gaussian (LoG), Morphological and Canny methods in all aspects with higher peak signal to noise ratio (PSNR) and signal to noise ratio (SNR) values. Hence, the proposed hybrid scheme is better and robust and results in accurate estimation of vehicle number from the blurry vehicle number plate (BVNP) images for the given datasets.

In this work, we are solving the major problem of reducing the time complexity of searching a string in huge corpus by using GPU as our computational environment (utilizing GPGPU and CUDA as programming platform) and Knuth–Morris–Pratt (KMP) and BMH (Boyer–Moore–Horspool) as string matching algorithms. String matching is a widely used technique in current research interest of various application areas such as bioinformatics, network intrusion detection, and computer virus scan. Although data are memorized in various ways, text remains the main form to exchange information. This is particularly evident in literature or linguistics where data are composed of huge corpus and dictionaries. These analytics are required in computer science where a large amount of data is stored in linear files. To search a particular string from these huge corpus takes more time in traditional CPU’s and this is a major problem.

Fractal dimension (FD) is a necessary aspect for characterizing the surface roughness and self-similarity of complex objects. However, fractal dimension gradually established its importance in the area of image processing. A number of algorithms for estimating fractal dimension of digital images have been reported in many literatures. However, different techniques lead to different results. Among them, the differential box-counting (DBC) was most popular and well-liked technique in digital domain. In this paper, we have presented an efficient differential box-counting mechanism for accurate estimation of FD with less fitting error as compared to existing methods like original DBC, relative DBC (RDBC), and improved box-counting (IBC) and improved DBC (IDBC). The experimental work is carried out by one set of fourteen Brodatz images. From this experimental result, we found that the proposed method performs best among the existing methods in terms of less fitting error.

Fractal dimension (FD) is most useful research topic in the field of fractal geometry to evaluate surface roughness of digital images by using the concept of self-similarity, and the FD value should lie between 2 and 3 for surfaces of digital images. In this regard, many researchers have contributed their efforts to estimate FD in the digital domain as reported in many kinds of the literature. The differential box-counting (DBC) method is a well-recognized and commonly used technique in this domain. However, based on the DBC approach, several modified versions of DBC have been presented like relative DBC (RDBC), improved box counting (IBC), improved DBC (IDBC). However, the accuracy of an algorithm for FD estimation is still a great challenge. This article presents an improved version of DBC algorithm by partitioning the box of grid into two asymmetric patterns for more precision box count and provides accurate estimation of FD with less fit error as well as less computational time as compared to existing method like DBC, relative DBC (RDBC), improved box counting (IBC), and improved DBC (IDBC).