Publication Search

Mobile face authentication for digital financial services must simultaneously satisfy recognition accuracy, computational efficiency, and biometric security under resource-constrained deployment conditions. This study proposes and evaluates a lightweight face-authentication framework that integrates detector–recognition pipeline optimization, protected biometric-template transformation, and blockchain-backed integrity support. Five face detection–recognition pipelines were systematically evaluated using a shared LightCNN-29v2 backbone fine-tuned on the Indonesian Muslim Student Face Dataset (IMSFD), with Mahalanobis-based Distance-Based Encryption (DBE) providing protected template matching and blockchain hash anchoring serving as an architectural integrity layer. Experiments on 3,660 images from 68 identities demonstrate that the MTCNN + LightCNN pipeline achieves the most favorable in-domain performance, reaching 94.95% accuracy, a ROC-AUC of 0.9970, an F1-score of 0.95, and successful processing of 3,546 out of 3,660 test images with an overall model size of approximately 5 MB. Applying Mahalanobis-based DBE further improves verification performance on IMSFD, increasing accuracy to 96.88% while reducing the False Acceptance Rate (FAR) from 0.83% to 0.18% and the False Rejection Rate (FRR) from 16.44% to 10.12%. Cross-dataset evaluation on LFW, CFP-FF, CFP-FP, AgeDB-30, CALFW, and CPLFW indicates that the proposed framework generalizes well to frontal-domain benchmarks but exhibits expected performance degradation under cross-pose and cross-age conditions due to distribution shift. Overall, the results demonstrate that detector selection is the dominant factor influencing end-to-end verification performance, while domain-specific fine-tuning and protected template matching are essential for secure and practical deployment in mobile financial authentication systems.

This study aims to analyze a comparison of coal stockpile volumes using the DJI Mavic 3 Pro Unmanned Aerial Vehicle (UAV) method versus the truck count method at PT. Mitra Barito. Data collection was conducted through aerial photography using a UAV at altitudes of 60 meters and 70 meters, as well as Ground Control Point (GCP) measurements using GPS. The aerial imagery data was processed using photogrammetry software to generate orthophotos and a Digital Elevation Model (DEM), followed by a geometric accuracy test based on the Geospatial Information Agency Regulation No. 6 of 2018, using the Circular Error 90% (CE90) and Linear Error 90% (LE90) parameters. The research results show that high-quality processing at an altitude of 60 meters yields a CE90 value of 2.1619 meters and an LE90 value of 4.3656 meters, thereby meeting the accuracy standards for RBI maps at a scale of 1:5,000, Class 3 for horizontal accuracy, and a scale of 1:10,000, Class 3 for vertical accuracy. Volume calculations of the stockpile using UAVs yielded a result of 22,750.900 m³, while the truck count method produced a volume of 23,503.300 m³. The volume difference between the two methods was 753.400 m³, with a deviation percentage of 3.2%. Based on the research results, the UAV method is considered capable of providing relatively accurate calculations of coal stockpile volume.

Gesture  recognition  technology  is  used  to  detect  movements  through  image processing,   enabling  computers  or digital devices to understand and interpret human  body  movements  as  input  or  commands.   This  technology  has  great potential  to bridge communication between the deaf community and individuals without   hearing   impairments,    enhancing  interaction  and  enriching  mutual understanding between the two.  However,  the accuracy ofgesture recognition is often  affected  by variations in the distance between hand landmarks.  Based on this problem,  this research proposes a methodfor stabilizing the measurement of distances between landmark points  in gesture recognition through a polynomial regression  approach.   Specifically,   the  distance  between  hand  landmarks  is calculated and stabilized using polynomial  regression to improve the accuracy of gesture recognition.  This method is implemented using the MediaPipeframework to detect and track hands in real-time,  and the OpenCV library to manage video. The  research  results  show  that  this  approach  can  significantly  improve  the stability  and accuracy  of gesture detection.   The developed system successfully detects gestures for  letters A  through F with a high accuracy  rate,  averaging above 98,3%.  The use ofpolynomial regression helps enhance detection accuracy by reducing noise in the landmark data.

Improper water management in rice cultivation can lead to water stress, which reduces productivity. Conventional monitoring has limitations on large-scale lands, necessitating more efficient remote sensing technologies. This study aims to develop a water stress identification system for rice plants in the late vegetative phase using multispectral drone imagery integrated with an Artificial neural network (ANN). The research method employs an experimental approach with six water availability levels in Karyamukti Village, Sumedang. Field reference data were obtained through soil moisture sensors converted into Available Water (AW) values. Image processing stages included orthomosaic reconstruction, leaf object segmentation, and transformation of vegetation indices (NDVI, NDRE, GNDVI, etc.) as model inputs. The results show that the ANN model with a four-hidden-layer architecture achieved training and validation accuracies of 94–95%. In the independent testing phase, the model produced an accuracy of 94.60% with an F1-Score of 93.33%. Spatial visualization of the prediction results indicates a consistent water condition distribution across rice plots. In conclusion, the integration of multispectral drones and ANN provides an accurate non-destructive solution for spatial monitoring of water availability in rice plants.

Digital images often experience noise disturbances that can reduce visual quality and interfere with the image analysis process. One common type of noise is salt and pepper noise, especially in grayscale images, which is characterized by the random appearance of black and white dots. This study applied the Deep Convolutional Autoencoder (DCAE) method with a skip connection mechanism to eliminate salt and pepper noise in grayscale images measuring 256×256 pixels. The dataset used consists of 300 pairs of clean images and noisy images that have gone through the preprocessing stage, including normalization and data augmentation. The model was trained using an Adam optimizer with a Mean Squared Error (MSE) loss function and validated through a train-test split scheme to avoid overfitting. Model performance was evaluated using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) metrics. The test results showed that the DCAE model with skip connections was able to effectively reduce noise while maintaining the main structure of the image based on the PSNR and SSIM values obtained, and showed better performance than conventional median filters. In addition, the model was successfully implemented into a Streamlit-based application to perform the image denoising process interactively, making it easier for users to experiment and visualize results in real-time.

In today’s digital era, social media has become a new space for individuals to develop their personal creativity. This study analyzes the content strategy of the TikTok account @buiramira in building audience engagement and understanding the impact it has on the account’s image. The background of this research stems from the phenomenon of increasing social media usage, particularly TikTok, as a means of communication, entertainment, and promotional media. Using McGuire’s information processing theory, the study finds that followers’ perceptions of the TikTok account @buiramira as a source of information for thesis learning are very positive, making it suitable as a learning platform. The content presented not only provides technical knowledge but also shapes students’ attitudes and behaviors in facing the thesis process.This research employs a descriptive qualitative method  with data collection techniques including content observation, interviews, and documentation. The results show that the @buiramira account successfully builds strong interaction with its audience through consistent uploads, the use of a communication style that feels close to viewers, and the utilization of current trends. In addition, the study finds that audience engagement is strongly influenced by a combination of content creativity, relevant topic selection, and the account owner’s ability to maintain emotional closeness with followers. These findings emphasize the importance of well-planned digital communication strategies in managing social media accounts, so that they can provide benefits both personally and professionally.

The increasing demand for low latency and high-throughput multimedia applications has spurred significant advancements in hardware software co design. This study explores the integration of custom digital signal processing (DSP) hardware accelerators with optimized software frameworks to enhance deep learning accelerated DSP tasks. The proposed co design approach significantly reduces latency and improves throughput compared to traditional software-only DSP implementations. Through the development of custom hardware accelerators built with FPGA technology, the system achieves up to a 1.85x reduction in latency and a 1.5x improvement in throughput for real-time multimedia tasks such as image recognition, video decoding, and audio processing. The combination of hardware and software optimizations allows for better resource utilization, enabling the parallel processing of computationally intensive tasks while the software framework handles less demanding operations. Additionally, the co design system demonstrated improved energy efficiency, making it highly suitable for embedded systems. The results show that the hardware software co design approach offers substantial advantages in performance, latency reduction, and energy efficiency, positioning it as a viable solution for real-time multimedia applications. The findings have important implications for applications requiring fast data processing, such as autonomous driving, healthcare, and disaster management. Future research could explore alternative hardware accelerators, advanced software optimizations, and AI-based resource management to further improve the system’s efficiency and scalability for more complex multimedia tasks.

Adulteration of rice bran is commonly done by mixing it with materials of similar appearance but lower nutritional value, such as ground rice husk. A key indicator of such adulteration is increased lignin content. Adding phloroglucinol solution to the mixture produces a red color that varies with lignin levels. This study aims to estimate lignin content in rice bran-husk mixtures using artificial intelligence and digital image processing. YCbCr color model images of eleven rice bran-husk compositions, treated with phloroglucinol, were analyzed. The lignin content of each variation was measured in the lab and used to define eleven classes. A Probabilistic Neural Network (PNN) was employed as the classifier, with image histograms of varying bin sizes as input. PNN performance was evaluated using 4-fold cross-validation. Results showed the highest average accuracy of 85.80% with 32 bins and histograms from all three YCbCr channels.