Deep learning for image processing and restoration

Image restoration has always been an ill-posed process due to the information loss. Some degradation can easily be simulated using mathematical formula. This simulation helps training data can be achieved at low cost. However, degradation as shadow is impossible to explicitly simulate by computer pr...

وصف كامل

محفوظ في:
التفاصيل البيبلوغرافية
المؤلف الرئيسي: Le, Ky Nam
مؤلفون آخرون: Wen Bihan
التنسيق: Final Year Project
اللغة:English
منشور في: Nanyang Technological University 2023
الموضوعات:
الوصول للمادة أونلاين:https://hdl.handle.net/10356/166645
الوسوم: إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
المؤسسة: Nanyang Technological University
اللغة: English
الوصف
الملخص:Image restoration has always been an ill-posed process due to the information loss. Some degradation can easily be simulated using mathematical formula. This simulation helps training data can be achieved at low cost. However, degradation as shadow is impossible to explicitly simulate by computer program. This makes dataset in the field of shadow removal become limited. This project aims to solve the shadow removal problem with low-cost dataset using deep learning methods. In this project, MaskshadowGAN unpaired shadow removal model is improved by equipping additional process to the original pipeline. Moreover, a method to attack BDRAR shadow detection model is discovered while experimenting a new unsupervised shadow removal pipeline. Finally, an application is developed for the users to interact with the shadow removal model.