International Journal of Signal and Imaging Systems Engineering (4 papers in press)

Regular Issues

• Understanding and analysis of mimicked speech: a case study  
  by Pallavi S. Marathe, Balasaheb J. Nagare 
  Abstract: Laboratory received an intriguing mimicry-based case. In that criminal case, defendant used his mimicked voice to negotiate a financial settlement. The studys goal was to determine whether the mimicked voice actually belonged to the target speaker or not. The study was built on the basis of a dialect, speaking style, and place of articulation. Since no comparable features could be derived using linear predictive analysis, phonetic features were thoroughly explored. The similarity between the accuseds mimicked speech and actual voice was proven by glottal leakage (B1). The jitter and shimmer calculations of the disturbances gave information about the accused and the target speakers ages. However, linguistic, phonetic, and acoustical research helped to bring the case to a successful conclusion. The mimicked voice, the accuseds natural voice, and the target speaker's voice could not be spectrographically matched.
  Keywords: mimicry; impersonation; formant frequency; spectrographic analysis; voice analysis.
  DOI: 10.1504/IJFE.2023.10053177
   
  
• Grape Cluster and Disease Detection with Hybrid Fuzzy Residual Maxout Network  
  by Rajkumar Bhimrao Pawar, Madhuri Rao 
  Abstract: This paper proposes a novel deep-learning method for grape cluster detection. Primarily, the grape cluster image is pre-processed using a bilateral filter to remove noise and the image is enhanced by the decorrelation stretching. Then, segmentation is performed using Mobile U-Net to segment the required regions. Later, features from the segmented images are extracted using different feature extractors. Afterwards, features fed into the Hybrid Fuzzy Residual Maxout Network (HFRMN) model for the detection of grape clusters. Here, the HFRMN model is designed by the incorporation of Deep Maxout Network (DMN), Deep Residual Network (DRN), and the Fuzzy concept. Finally, disease detection is accomplished by utilizing the proposed HFRMN. Moreover, HFRMN attained superior values of 90.0% for True Positive Rate (TPR), 91.3% for True Negative Rate (TNR), 91.4% for accuracy, 92.4% for F1-score, 71.3% for computational efficiency, 1.04% for Mean Squared Error (MSE), and 89.6% for Mean Average Precision (MAP).
  Keywords: grape cluster; bilateral filter; decorrelation stretching; Deep Maxout Network; Deep Residual Network.
  DOI: 10.1504/IJSISE.2024.10069173
   
  
• A Systematic Analysis of Intelligent Control for Robotic Arms Using Non-Invasive EEG Signals: a Comprehensive Review  
  by Senthil Vadivelan D, Uma M, Prabhu Sethuramalingam 
  Abstract: The Brain-Computer Interface (BCI) technology has evolved into a powerful tool for human-machine interaction, particularly benefiting those with physical limitations. This review focuses on the significance of non-invasive electroencephalography (EEG) signals in BCI applications, emphasizing their role in controlling assistive robotic arms. The manuscript explores established techniques for processing electrophysiological data, feature extraction, and the use of classification algorithms. Also, discuss diverse BCI hardware for comprehensive brain signal acquisition analysis. The paper addresses challenges in BCI assistive robotic arm control applications and suggests potential solutions, serving as a valuable resource for researchers. Furthermore, it identifies research gaps, helping to understand and resolve emerging issues in BCI technology and assistive robotic arm control.
  Keywords: Brain-Computer Interface (BCI); EEG Signal; Robot arm control; Feature extraction; Machine Learning.
  DOI: 10.1504/IJSISE.2024.10070154
   
  
• SC-PCA-Enr: Multi-Modal Residual Adaption Enabled ShearletContourlet Autoencoder for Fused Multi Modal Image Enhancement  
  by Nitin Sudhakarrao Thakare, Mukesh Yadav 
  Abstract: Image fusion involves combining redundant and complementary data from multiple source images, but creating strong features and discriminating models is challenging. The research proposes a fusion-based shearlet counterlet-enabled PCA-auto encode image enhancer (SC-PCA-EnR enhancer) to improve image quality by producing high contrast in bright areas and improving visibility in dark areas. The Non-Subsampled Shearlet Contourlet Transform (NSSCT) is used for multi-resolution and multi-directional analysis, while Principal Component Analysis reduces dimensionality and enhances interpretability. The use of an auto encoder based image enhancer captures the intricate patterns and structures in images by automatically learning features from the input image. The SC-PCA-EnR enhancer modal outperforms traditional methods in performance metrics such as Similarity index measure (SSIM) of 0.97, Peak Signal Noise ratio (PSNR) of 51.98 dB, and Mean Square Error (MSE) of 0.57, and normal brain dataset.
  Keywords: Non-Sub sampled Shearlet Contourlet transforms; multimodal fusion; image enhancement; Principal component analysis; autoencoder-based image enhancer.
  DOI: 10.1504/IJSISE.2024.10070715