As an answer to the need to prove the authenticity of digital images, several forensic tools for image forgery localization have been proposed in the past. However, authenticity analysis remained a challenging task, which requires an expert's knowledge to correctly interpret each forensic algorithm's output. Moreover, since different tools look for different manipulation traces, a thorough analysis requires the joint interpretation of the maps produced by several tools, which is nontrivial since each tool's reliability is possibly affected by different elements. Recently, deep learning-based forgery localization schemes were proposed, allowing for more automated reasoning; however, their accuracy significantly decreases when they are tested on forgeries that deviate from those used for the training phase. This work proposes a deep learning-based framework that merges the forgery localization maps provided by model-based image forensics tools based on the U-Net architecture. The experiments show that the proposed approach improves the quality of forgery localization maps compared to those produced by single tools and by state-of-the-art fusion frameworks while simultaneously achieving a desirable generalization capability.
Pengpeng Yang, Daniele Baracchi, Dasara Shullani et al.· Frontiers in Imaging· 0 citations
Quantitative precipitation estimation (QPE) is a fundamental task of hydrometeorological applications, ranging from flash-flood detection to water-resource management. While weather radars offer superior spatial coverage compared with rain gauges, traditional estimation based on the empirical Z−R (reflectivity factor vs. rainfall rate) relationship fails to capture spatial variability and tends to underestimate extreme rainfall. The idea pursued here is to learn the radar-to-rainfall mapping by means of a convolutional neural network (CNN) trained on co-located radar and rain-gauge data; thus, once trained, the network can convert radar reflectivity measures into rainfall values, even where gauges are unavailable. Although machine learning techniques are promising for this task, they typically demand large training sets. To operate in a limited-data regime, we introduce a U-Net architecture that separates the analysis of space from that of time: each block first looks at the structure of the reflectivity field and then at how it changes over consecutive radar scans, extracting spatiotemporal features from volumetric data with fewer parameters than a full three-dimensional filter. The model is evaluated on a severe convective event that affected Tuscany, Italy, on 2 November 2023, benchmarking its performance against the classical Joss–Waldvogel Z−R relationship (suitable for convective events), a data-driven log-regression of weather-radar and rain-gauge data, and a baseline CNN architecture. The main advantages are negative bias—i.e., underestimation of rainfall—more than halved and correlation with rain-gauge measures more than doubled, under the same operational conditions. What is noteworthy is the capability of learning the model from radar and rainfall data taken in different times and places, as well as the possibility of converting a reflectivity map into a rainfall map without the need for simultaneous rain-gauge measures. This is an asset of fixed parametric methods; however, they are far less accurate.
Matteo Passeri, F. Argenti, Daniele Baracchi et al.· Environments· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
