ResearchBib Share Your Research, Maximize Your Social Impacts
Sign for Notice Everyday Sign up >> Login


Journal: International Journal of Advanced Research in Engineering and Technology (IJARET) (Vol.12, No. 03)

Publication Date:

Authors : ;

Page : 421-429

Keywords : Matting; Motion cues; Soft Segmentation; GAN.;

Source : Downloadexternal Find it from : Google Scholarexternal


Since the Image processing term was invented, scientists were building networks that simplifies object extraction process, and binding those objects into existing images. At the beginning the other algorithms were enough to do this operation, but after the high-quality cameras has made an appearance with a high-resolution image, those algorithms have become weak to find and also extract objects, especially when there is just a slight deference between the background and foreground colors. As an attempt to solve that, we introduce a two-deep learning-based models. The first one, tries to find the location of the target object using supervised network encoder-decoder principle, it's output will be feeded to” the second one which is” a generative adversarial network that tries to improve the predictions and has more accurate results. After testing, this combination of two models have proved itself as a more efficient way to solve the matting problem. After gaining the almost-perfect results of this solution from an image an entire human body in it, the next step will be to take a pre-refined image of a piece of clothes and compose it on the consumer body to give him an experience of seeing how it will look on him before even he purchases it.

Last modified: 2021-03-30 15:47:29