IJIGSP Vol. 15, No. 3, Jun. 2023
Cover page and Table of Contents: PDF (size: 669KB)
This work introduces the novelty as an application of histogram-based bins approach with statistical moments for detecting and classifying malaria using blood smear images into parasitized and uninfected cell images and the rising disease of COVID-19/Normal lung images. Proposed algorithms greatly vary as compared to the previous work. This work aims to improve accuracy in detection and classification and reduce feature vector dimensionality. It focuses on detailed image contents extracted into 8 bins by considering the significance of the R, G, and B color component relationship in the formation of each pixel. The texture features are represented by the first four moments for each of the three colors separately. This leads to the generation of 12 features vectors, each of size 8 components for each image in the database. Feature dimensionality reduction is achieved by applying different feature selection techniques to obtain desired optimum feature space. The comprehensive feature analysis presented here identifies many useful findings in order to validate the contribution of each image content uniquely in detection and classification. The proposed approach experimented with two image datasets: the malaria dataset obtained from the National Library of Medicine (NLM) and the lung image dataset acquired from the Radiography Database from Kaggle. The performance of work presented here is evaluated and compared with previous work with the same set of parameters, namely precision, recall, F1 score, and the AUC. We have achieved and improved the performances compared to previous work and also achieved better results even for the COVID-19 dataset.[...] Read more.
Video steganography is used to conserve the confidential information in various security applications. To give advance protection to the secrete message, pixels locations are optimized using nature inspired algorithm. The input video is separated into a sequence of still image frames then key frames are extracted. The proposed Required Pixel Density (RPD) value calculation and feature extraction are carried out on the extracted frames to perform the frame classification. The frame classification is done using proposed Fractional Water-Earth Worm optimization algorithm based Deep Convolutional Neural Network (FrWEWO-Deep CNN) in order to classify the frames as high, low and medium quality. Thus pixel location prediction is carried out using trained Deep CNN then secret image is hide within high quality frame with Wavelet Transform (WT) and Inverse WT (IWT). Peak Signal to Noise Ratio (PSNR) and Correlation Coefficient (CC) are performance evaluation parameters. For efficient video steganography better imperceptibility and robustness are required. Imperceptibility is a scale of PSNR value showing similarity between original and stego video frames. The robustness of video steganography is measured by CC between embedded and extracted secret images. The proposed algorithm gives enhanced performance is compared with previous state of art such as WEWO-Deep RNN. The PSNR value is progressed from 41.8492 to 46.5728 dB and CC value improved from 0.9660 to 0.9847.[...] Read more.
Ultrasound is mostly used for diagnosis to deal with the specific abnormality in human body. To observe the internal organs including liver, kidneys, pancreas, thyroid gland, ovaries etc. ultrasound can be used. In diagnostic applications, 2 to 18 MHz frequencies are used. The sound wave explorations occurred through soft tissue and fluids. It bounces back as echoes from denser surfaces and creates an image. While producing ultrasound images from echo signal speckle noise is induced in a multiplicative way. Thus, speckle becomes the key challenge for ultrasound imaging. Several speckle reducing linear, non-linear and anisotropic diffusion-based methods are implemented to preserve the sharp edges of ultrasound images. Those methods contain lake of smoothing and edge preservation. However, this research proposed a combined method of adaptive filter (wiener) and anisotropic diffusion (modified Perona Malik) for speckle reduction of 2D ultrasound images by retain the important anatomical features. A comparison of all the existing methods studied based on the simulated experiment. To test the methods liver, kidney, heart and pancreas noise free images are used. Then, speckle noise is manually added with distinguished variance in between 0.02 and 0.20. Quality metrics are used to test the performance and show the improvements of the proposed method. About 71.79% structure similarity (SSIM), 66.72% root mean square error (RMSE), 56.93% signal to noise ratio (SNR), and 62.30% computational time are improved on average compared with the other methods.[...] Read more.
Agriculture is a big sector in nations like India, and it provides a living for many people. To improve crop productivity, it’s very necessary to identify and classify plant diseases and prevent them from spreading further so that they do not affect the whole plant. Artificial intelligence (AI) and computer vision can help detect plant diseases that humans cannot always catch and overcome the shortcomings of continuous human monitoring. In this article, we aim to detect and classify diseases in tomato and apple leaves using deep learning approaches and compare the results between different models. Because tomatoes and apples are important components of the human diet, crop waste can result in losses for both farmers and ordinary people. These plant diseases have an immediate and negative impact on both the amount and quality of yield. Crop diseases must be identified and prevented as soon as possible to improve crop yield. Therefore, we need to monitor and analyze the growth stages of the plants so that the farmers can produce disease-free and with minimal losses to the crop. Furthermore, we used the sequential convolutional neural network (CNN) model followed by transfer learning models like VGG19, Resnet152V2, Inception V3, and MobileNet and compared the models based on accuracy. The performance of the models was evaluated using various factors such as dropout, batch size, and the number of epochs. For both, the datasets, the tomato, and apple MobileNet architecture performed better than the other existing models.[...] Read more.
Wavelet transform has become a popular tool for signal denoising due to its ability to analyze signals effectively in both time and frequency domains. This is important because the information that is not visible in the time domain can be seen in the frequency domain. However, there are many wavelet families and thresholding techniques (such as haar, Daubechies, symlets, coiflets, meyer Gaussian, morlet, etc) thatare available for the analysis of signals, and choosing the best out of them all is usually time-consuming, thus making it a difficult task for researchers. In this article, we proposed and applied a stepwise expository-based approach to identify the wavelet family and thresholding technique using real-time signal power data acquired from Long-Term Evolution (LTE). We found out from the results that Rigrsure thresholding with the Daubenchies family outperforms others when engaged in practical signal processing. The stepwise expository-based approach will be a relevant guide to effective signal processing over cellular networks, globally. For validation, different datasets were used for the analysis and Rigrsure outperforms the other thresholding techniques.[...] Read more.
This paper projects, the impact & accuracy of speech compression on AER systems. The effects of various codecs like MP3, Speex, and Adaptive multi-rate(NB & WB) are compared with the uncompressed speech signal. Loudness enlistment, or a steeper-than-normal increase in perceived loudness with presentation level, is associated with sensorineural hearing loss. Amplitude compression is frequently used to compensate for this abnormality, such as in a hearing aid. As an alternative, one may enlarge these by methods of expansion as speech intelligibility has been represented as the perception of rapid energy changes, may make communication more understandable. However, even if these signal-processing methods improve speech understanding, their design and implementation may be constrained by insufficient sound quality. Therefore, syllabic compression and temporal envelope expansion were assessed for in speech intelligibility and sound quality. An adaptive technique based on brief, commonplace words either in noise or with another speaker competing was used to assess the speech intelligibility. Speech intelligibility was tested in steady-state noise with a single competing speaker using everyday sentences. The sound quality of four artistic excerpts and quiet speech was evaluated using a rating scale. With a state-of-art, spectral error, compression error ratio, and human labeling effects, The experiments are carried out using the Telugu dataset and well-known EMO-DB. The results showed that all speech compression techniques resulted in reduce of emotion recognition accuracy. It is observed that human labeling has better recognition accuracy. For high compression, it is advised to use the overall mean of the unweighted average recall for the AMR-WB and SPEEX codecs with 6.6 bit rates to provide the optimum quality for data storage.[...] Read more.
This work presents an algorithm for the automatic detection of cardiomegaly on CXR images. Cardiomegaly is a medical condition in which the heart becomes enlarged than the actual and the efficiency of the heart would decrease and sometimes congestive heart failure occurs. Although there could be numerous reasons, high blood pressure and coronary artery disease are the main causes of cardiomegaly. Hence, the main intention of this work is to develop a CNN based model to efficiently identify the presence of cardiomegaly abnormality. The learning phase of the model is achieved by using CXRs that are extracted from the publically available “chest x-ray14” medical dataset and to compute the proposed model performance, an experimental platform is designed and implemented in the MATLAB tool. We have trained the model with 100, 120, 150, and 200 epochs. But the trained model with 120 epochs shows a revolutionary outcome. The acquired accuracies of 100,120,150 and 200 epochs are 84.69%, 98.00%, 89.09% and 87.64% respectively. However, many approaches have been developed for cardiomegaly identification but the proposed model shows record performance.[...] Read more.