Precipitation Processes
Precipitation Gravimetry
Precipitation and Co-precipitation
Uniform Depth Channel Flow: Problem Solving
Uniform Depth Channel Flow
Precipitation Titration: Endpoint Detection Methods
You might also read
Articles linked to this work by shared authors, journal, and citation graph.
Updated: Nov 22, 2025

Test Samples for Optimizing STORM Super-Resolution Microscopy
Published on: September 6, 2013
Rain can make it hard to see in surveillance and monitoring systems. This study introduces a new method for removing rain from images by considering how depth affects visibility. The researchers developed a model that accounts for both rain streaks and fog. They also created a dataset of real outdoor photos to train their method. The proposed approach uses a deep neural network that learns depth-guided features and produces a clean image in real-time. The method outperforms existing techniques in tests. The results suggest that depth is an important factor in improving rain removal performance.
Area of Science:
Background:
Environmental visibility is often reduced by rain, affecting surveillance and monitoring systems. Rain streaks and fog obscure distant objects more than close ones. This depth-dependent effect is not widely considered in current rain removal techniques. Existing datasets and approaches fail to incorporate physical rain properties. As a result, real-world rain removal remains inefficient. Prior research has shown that depth influences visibility. However, no prior work had resolved how to model this in rain removal. This gap motivated the development of a new imaging model. The study aims to address this limitation through a novel approach.
Purpose Of The Study:
This study aims to improve rain removal by incorporating scene depth into the imaging model. The researchers propose a new method that considers both rain streaks and fog. They aim to create a dataset that reflects real-world conditions. The goal is to design a deep neural network for real-time processing. The model should learn depth-guided non-local features. The researchers also seek to regress a residual map for clean images. Their approach is intended to outperform existing techniques. The study tests the method against state-of-the-art alternatives.
Main Methods:
The researchers developed a new rain imaging model that includes depth effects. They created a dataset called RainCityscapes using real outdoor photos. The model combines rain streaks and fog into a unified framework. A deep neural network was designed for real-time processing. The network learns depth-guided non-local features. It regresses a residual map to produce a rain-free image. The method was trained on the RainCityscapes dataset. The researchers tested the model against other leading techniques.
Main Results:
The proposed model outperformed existing methods in both visual and quantitative tests. The RainCityscapes dataset provided realistic training and evaluation data. Depth-guided non-local features improved rain removal accuracy. The residual map effectively reduced rain streaks and fog. The model achieved real-time performance on standard hardware. It showed better results on distant objects than on closer ones. The method maintained high image quality and detail. The researchers demonstrated the model's superiority in multiple experiments.
Conclusions:
The study shows that incorporating depth into the rain model improves removal performance. The proposed method outperforms others in real-world scenarios. The RainCityscapes dataset supports more accurate training. Depth-guided non-local features enhance the model's effectiveness. The residual map contributes to better image clarity. The model processes images in real-time, making it practical for use. The results suggest that depth is a key factor in rain removal. The authors propose that this approach advances the field of image restoration.
The method incorporates depth-guided non-local features to improve rain removal accuracy.
It uses real outdoor photos to reflect realistic rain conditions and depth effects.
Depth affects how rain and fog obscure objects, with distant objects being more affected.
The residual map helps produce a clean image by removing rain streaks and fog.
The deep neural network is optimized for fast processing on standard hardware.
They propose that depth is a key factor in improving rain removal accuracy.