This website requires JavaScript.

Progressive Update Guided Interdependent Networks for Single Image Dehazing

Aupendu KarSobhan Kanti DharaDebashis SenPrabir Kumar Biswas
摘要
Images with haze of different varieties often pose a significant challenge todehazing. Therefore, guidance by estimates of haze parameters related to itsvariety would be beneficial and they should be progressively updated along withiterative haze reduction to allow optimal dehazing. To this end, we propose amulti-network dehazing framework containing novel interdependent dehazing andhaze parameter updater networks that operate within a unique iterativemechanism. The haze parameters, transmission map and atmospheric light, arefirst estimated using specific convolutional networks allowing color casthandling. The estimated parameters are then used as priors in our dehazingmodule, where the estimates are progressively updated by novel convolutionalnetworks using the iterative mechanism. The updating takes place jointly withprogressive dehazing by a convolutional network that invokes inter-iterationdependencies. The joint updating and dehazing within the iterative mechanismgradually modify the haze parameter estimates toward achieving optimaldehazing. Through ablation studies, our iterative dehazing framework is shownto be more effective than the use of conventional LSTM based recurrence,image-to-image mapping and haze model based estimation. Our dehazing frameworkis qualitatively and quantitatively found to outperform the state-of-the-art onsynthetic and real-world hazy images of several datasets with varied hazyconditions.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答