Browsing by Author "Lim, Long ang"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Ön plan bölütlenmesinde denetimli çok ölçekli konvolüsyonel sinir ağları yaklaşımının kullanımı
    (Fen Bilimleri Enstitüsü, 2018) Lim, Long ang; Keleş, Hacer; Bilgisayar Mühendisliği
    Several methods have been proposed in foreground segmentation domain. However, they lack the ability of handling various difficult scenarios such as illumination changes, background or camera motion, camouflage effect, shadow etc. To address these issues, we propose three different robust encoder-decoder type deep neural networks that can be trained end-to-end using only a few training examples; first, we adapt a pre-trained convolutional network, i.e. VGG-16 Net, under a triplet framework in the encoder part to embed an image in multiple scales into the feature space and use a transposed convolutional network in the decoder part to learn a mapping from feature space to image space. Second, we propose a Feature Pooling Module (FPM) that can be plugged on top of a single input encoder to extract multiple scale features and the same decoder is embedded on top of these features to learn upsampling to image space. Third, we extend the FPM module by introducing features fusion inside this module, resulting in a robust module against camera motions and we further propose a novel decoder network on top of the extended FPM for further performance improvement. In order to evaluate our models, we entered the Change Detection 2014 Challenge (changedetection.net) and our methods called FgSegNet_M, FgSegNet_S and FgSegNet_v2 outperformed all the existing state-of-the-art methods by an average F-Measure of 0.9770, 0.9804 and 0.9847, respectively. We also evaluate our models on SBI2015 and UCSD Background Subtraction datasets. In the context of this study, we also provide a comprehensive study about patch-wise learning in foreground segmentation domain. Furthermore, in order to evaluate the methods that we developed in the context of the foreground segmentation problem in semantic segmentation domain, we present two semantic segmentation method studies in detail.