The rapidly developed Health 2.0 technology has provide people with more opportunities to conduct online medical consultation than ever before. Understanding contexts within different online medical activities becomes a significant issue to facilitate patients' medical decision making process. As a subcategory of machine learning, neural networks have drawn increasing attentions in natural language processing applications. In this study, we focus on the modeling and analysis of patient-physician-generated data based on an integrated CNN-RNN framework. A CNN-based classifier is designed to extract textural features in the sentence-level, and a RNN-based model is constructed to learn the potential patterns and correlations in a dialog-level. An intelligent recommendation mechanism is then developed to provide patients with automatic clinic guide and pre-diagnosis suggestions in a data-driven way. Experiments based on the collected real world data demonstrate the effectiveness of our proposed model and method for online medical pre-diagnosis support.
Format-compatible encryption can be used to ensure the security and privacy of JPEG images. Recently, a JPEG image encryption method proved to be secure against known plaintext attacks by employing an adaptive encryption key, which depends on the histogram of the number of non-zero alternating current coefficients (ACC) in Discrete Cosine Transform (DCT) blocks. However, this scheme has been demonstrated to be vulnerable to chosen-plaintext attacks (CPA) based on the run consistency of MCUs (RCM) between the original image and the encrypted image. In this paper, an improved CPA scheme is proposed. The method of incrementing run-length values instead of permutation is utilized to satisfy the uniqueness of run sequences of different minimum coded units (MCUs). The experimental results show that the proposed method can successfully recover the outlines of plaintext images from the encrypted images, even with lower-quality factors.
28 January 2025 in Multimedia Tools and Applications
Singing melody extraction from polyphonic music is a complex but important task in music information retrieval. Most existing models struggle with adaptive multi-scale feature extraction and long-range dependency modeling, often resulting in common artifacts such as octave errors and pitch fragmentation. To address this, we propose the Cross-Scale Transformer Network (CSTNet), a novel deep learning architecture that dynamically captures and fuses multi-scale temporal and spectral features. The CSTNet introduces two key components: the Multi scale Time-Frequency Aggregation (MF-TFA) module for adaptive local feature representation across resolutions, and Feature Cross Transformer Block (FCTB), which enables explicit cross scale interaction through a global channel-wise cross-attention mechanism. Experimental results show that our proposed method outperforms five compared state-of-the-art methods, achieving overall accuracy scores of 87.3%, 90.5%, and 77.4% on the ADC 2004, MIREX 05, and MEDLEY DB datasets, respectively. Visualization results demonstrate that CSTNet effectively reduces octave errors.