46 research outputs found
Reflection removal and feature extraction techniques in non-cooperative visible eye images for iris recognition
Get PDFResearch on iris recognition system nowadays focuses on identifying a person in a non-cooperative environment by capturing an eye image in motion and at different distances. A visible wavelength illumination is used to capture the eye image which is believed to be safer to the eyes as excessive level of near infrared wavelength illumination can endanger the eye. However, the quality of data captured is very low and there are large reflections with different intensities in the eye image. These have caused incorrect segmentation of iris boundaries as well as the inability to extract texture features of an iris in a non-cooperative environment leading to a reduction in the iris recognition performance. The research proposed the development of two combined methods to improve the iris recognition system. The first combined method consists of line intensity profile and support vector machine and the second is a combination of multiscale sparse representation of local Radon transform. The former identifies and classifies between reflections and nonreflections whereas the latter performs three processes: reduces noise during down sample of normalized iris, extracts an iris texture in the different angles of orientation information and uses score combination of multiscale at the end of the process to increase the matching score. These two methods were tested against UBIRIS.v2 iris database and the results of iris recognition compared to the existing methods achieved an accuracy of more than 90%
Email spam classification based on deep learning methods: A review
Get PDFEmail spam is a significant issue confronting both email consumers and providers. The evolution of spam filtering has progressed considerably, transitioning from basic rule-based filters to more sophisticated machine learning algorithms. Deep learning has become a potent collection of techniques for addressing intricate issues such as spam classification in recent times. A thorough literature evaluation is required to have a comprehensive overview of the current research on utilizing deep learning methods for email spam classification. This review aims to identify the various deep learning techniques used for email spam, their effectiveness, and areas for future research. By synthesizing the outcomes of pertinent studies, this review delineates the strengths and drawbacks of various approaches, offering valuable insights into the challenges that must be tackled to enhance the precision and efficacy of email spam classification
Dual image watermarking based on NSST-LWT-DCT for color image
Get PDFAdvanced internet technology allows unauthorized individuals to modify and distribute digital images. Image watermarking is a popular solution for copyright protection and ensuring digital security. This research presents an embedding scheme with a set of conditions using non-subsampled Shearlet transform (NSST), lifting wavelet transform (LWT), and discrete cosine transform (DCT). Red and green channels are employed for the embedding process. The red channel is converted by NSST-LWT. The low-frequency area (LL) frequency is then split into small blocks of 8×8, each partition block is then transformed by DCT. The DCT coefficient of (3,4), (5,2), (5,3), (3,5), called matrix M1, and (2,5), (4,3), (6,2), (4,4), called matrix M2 are selected for singular value decomposition (SVD) process. With a set of conditions, the watermark bits are incorporated into those singular values. The green channel is cropped to get the center image before splitting into 4×4 pixels. The block components are then selected based on the least entropy value for the embedding regions. The selected blocks are then computed using LWT-SVD. A set of conditions for U(1,1) and U(2,1) are used to incorporate the watermark logo. The experimental findings reveal that the suggested scheme achieves high imperceptibility and resilience under various evaluating attacks with an average peak signal-to-noise ratio (PSNR) and correlation value (NC) values are up to 43.89 dB and 0.96, respectively
Smart object detection using deep learning algorithm and jetson nano for blind people
Get PDFVisual impairment is often defined as a best corrected visual acuity of worse than either 20/40 or 20/60. The term blindness is used for complete or nearly complete vision loss. Visual impairment may cause people difficulties with normal daily activities such as driving, reading, socializing and walking. Therefore, this project develops a smart object detection using deep learning algorithm and jetson nano to improve object detection for blind people. Furthermore, an ultrasonic sensor and sound notification are added to the project development to notify blind people if any object located nearest around them
Dual image watermarking based on NSST-LWT-DCT for color image
Get PDFAdvanced internet technology allows unauthorized individuals to modify and distribute digital images. Image watermarking is a popular solution for copyright protection and ensuring digital security. This research presents an embedding scheme with a set of conditions using non-subsampled Shearlet transform (NSST), lifting wavelet transform (LWT), and discrete cosine transform (DCT). Red and green channels are employed for the embedding process. The red channel is converted by NSST-LWT. The low-frequency area (LL) frequency is then split into small blocks of 8×8, each partition block is then transformed by DCT. The DCT coefficient of (3,4), (5,2), (5,3), (3,5), called matrix M1, and (2,5), (4,3), (6,2), (4,4), called matrix M2 are selected for singular value decomposition (SVD) process. With a set of conditions, the watermark bits are incorporated into those singular values. The green channel is cropped to get the center image before splitting into 4×4 pixels. The block components are then selected based on the least entropy value for the embedding regions. The selected blocks are then computed using LWT-SVD. A set of conditions for U(1,1) and U(2,1) are used to incorporate the watermark logo. The experimental findings reveal that the suggested scheme achieves high imperceptibility and resilience under various evaluating attacks with an average peak signal-to-noise ratio (PSNR) and correlation value (NC) values are up to 43.89 dB and 0.96, respectively
Generative Adversarial Networks In Object Detection: A Systematic Literature Review
Get PDFThe intersection of Generative Adversarial Networks (GANs) and object detection represents one of the most promising developments in modern computer vision, offering innovative solutions to longstanding challenges in visual recognition systems. This review presents a systematic analysis of how GANs are transforming these challenges, examining their applications from 2020 to 2025. The paper investigates three primary domains where GANs have demonstrated remarkable potential: data augmentation for addressing data scarcity, occlusion handling techniques designed to manage visually obstructed objects, and enhancement methods specifically focused on improving small object detection performance. Analysis reveals significant performance improvements resulting from these GAN applications: data augmentation methods consistently boost detection metrics such as mAP and F1-score on scarce datasets, occlusion handling techniques successfully reconstruct hidden features with high PSNR and SSIM values, and small object detection techniques increase detection accuracy by up to 10% Average Precision in some studies. Collectively, these findings demonstrate how GANs, integrated with modern detectors, are greatly advancing object detection capabilities. Despite this progress, persistent challenges including computational cost and training stability remain. By critically analyzing these advancements and limitations, this paper provides crucial insights into the current state and potential future developments of GAN-based object detection systems
Real-Time Object Detection System for Hospital Assets Using YOLOv8
Get PDFHospital administration is essential in the provision of high-quality social services to patients. Hospitals must have efficient asset management to offer quality medical care. On the other hand, many hospitals face problems such as data entry errors. Based on this problem, the author hopes to solve it by implementing real-time object detection and recording data distribution using the YOLO (You Only Look Once) algorithm. This data distribution will then be applied to the current system. Performance tests were carried out in this research using the YOLO architecture, especially on YOLOv8. one of the improvements of popular deep learning algorithms. This research used 7680 images (augmentation) which were divided into 3 parts. 6720 training data (88%), 640 validation data (8%), and 320 (4%) test data. 7680 data were added from 16 tested medical device categories with 200 images per category. This research has an average accuracy of 90%, an average precision of 94%, and an average recall value of 92.2%. These results show that YOLOv8 performs well in detecting medical devices. To improve accuracy, it is recommended to test larger and more diverse datasets. This research helps the healthcare industry better monitor and manage real-time assets
A low lighting or contrast ratio visible iris recognition using iso-contrast limited adaptive histogram equalization
No full textCondition of eye images with a low lighting or low contrast ratio between the iris and pupil is one of the challenges for iris recognition in a non-cooperative environment and under visible wavelength illumination. Incorrect iris localization can affect the performance of the iris recognition system. Iso-contrast limited adaptive histogram equalization is proposed to overcome this challenge and increase the performance of iris localization. The eye image is partitioned into the contextual sub-region; then, the proposed method transfers the pixel intensity by referring to a local intensity histogram and a newly suggested cumulative distribution function. This research was tested on 1000 eye images from the UBIRIS.v2 dataset. The results showed that the proposed method performed better than existing methods when dealing with a low lighting or low contrast ratio between the iris and pupil in the eye image
Analysis on texture and colour based features of periocular for low resolution colour iris images
Get PDFThe low resolution iris images in non-cooperative environment has resultant in failure to determine the eye center, limbic and pupillary boundary of the iris segmentation. Hence, a combination with periocular area is suggested to improve the accuracy of the recognition system. However, the existing periocular features extraction methods to extract the texture features can be easily affected by a background complication and depends on image size and orientation. Although some of the existing studies have combined the texture and colour features to increase the accuracy of periocular recognition, still, the method of colour feature extraction is limited to spatial information and quantization effects. This paper presents the analysis of texture and colour based features of periocular for low resolution colour iris images. Two datasets: UBIRIS.v2 and UBIPr are used and the proposed method provides robust discriminative structure features and sufficient spatial information which has increased the discriminating power
Template matching analysis using neural network for mobile iris recognition system
Get PDFToday, the use of mobile phone among individuals with an advanced technology, acquisition and computation resources is crucial in securing personal data and services as well as in identifying a person. Recently, there has been growing interest of researchers in iris recognition using mobile as an acquisition device to capture an eye image in a non-cooperative environment (static motion and at different distances). However, the use of mobile as an acquisition device still facing some noise factors because the quality of image captured in this environmental condition is low compared to a camera device. Besides, low awareness of user in handling the mobile devices and lack technical experiences in capturing iris images with this environmental condition is generally uncontrolled contribute to producing unpredictable low quality of eye images. Due to these issues, it led to incorrect segmentation of iris boundaries and subsequently lowered the ability to match the iris features. Hence, this condition contributes to performance degradation of iris recognition system. Therefore, to improve the ability to match the iris features in mobile iris recognition system, a combination with neural network is proposed. The iris database used to test against these methods is CASIA-Iris-M1-S2. The proposed method, a combination of Hamming distance and neural network has achieved a better result which 98.77% in term of accuracy for mobile iris recognition under non-cooperative environment
