000			01758nam a2200145 4500
005			20251101124006.0
008			251101b |||||||| |||| 00| 0 eng d
100			_aBalla Pavan Kumar
245			_aMF-MSCNN: Multi-Feature based Multi-Scale Convolutional Neural Network for Image Dehazing via Input Transformation
300			_ap1547-1559
520			_aRecently, many deep learning algorithms have been proposed for image dehazing. However, in most of these techniques, the issues of under-exposure and over-saturation are observed. These problems appear because of inadequate consideration of the overall haze level, i.e. Less-Haze (LH), Medium-Haze (MH), and High-Haze (HH), of hazy images while dehazing. Therefore, a Multi-Feature-based Multi-Scale Convolutional Neural Network (MF-MSCNN) is proposed, which considers the haze density of hazy images as a parameter while dehazing. Firstly, the classification operation is performed to categorize the hazy images into LH, MH, and HH. Based on this categorization, a haze density map is generated, which is concatenated to the input hazy image as part of input transformation (IT). Subsequently, the LH, MH, and HH features are extracted using the proposed MF-MSCNN. These features are adaptively chosen by the pooling layer to obtain an efficient transmission map from which the dehazed image is retrieved. The proposed work using IT operation and the MF-MSCNN model produces better results for all categories of hazy images when compared to the existing methods
654			_aImage dehazing _aDeep learning _aMultiple feature extraction _aHaze density map _aConvolutional neural network _aInput transformation
773	0		_080269 _9114212 _dNew Delhi IETE _tIETE Journal of Research _x0377-2063
942			_cJA
999			_c133159 _d133159