The evolution and progress of mobile devices and social networks have made it a simple task for inexperienced users to acquire photos and share them on social platforms. Due to the availability of media editing softwares, forgers can efficiently spread rumors thereby creating forged images/ videos (i.e., manipulated/ tampered content) which may create negative sentiments among the people of the country. Thus, the availability of highly sophisticated systems/ techniques for finding tampered traces in images and videos has received significant attention. With the availability of the conventional methods for forgery detection, the detection of the forgery in all the categories of images is not possible especially when the manipulated images are heavily compressed. This paper highlights and analyses the challenges of conventional forensic methods for forgery localization in heavily compressed tampered images. To meet the specific necessities and challenges of forgery localization in images, in this paper we investigated the encoder-decoder based deep learning frameworks (i.e., SegNet and U-Net) for segmentation of the forged regions from the holistic tampered images. Also, we have investigated the perception capabilities of the backbone CNN architectures on the encoder-decoder frameworks for precise segmentation of the forged regions from the images. Experiments have been conducted on publicly available datasets including CASIA-V2 and CMFD datasets and our own designed dataset. Experimental results demonstrate that SegNet+MobileNet on CASIA-V2 dataset, U-Net+AlexNet on CMFD dataset, and SegNet+ResNet-101 have been observed to precisely extract the forged portions from the entire altered images with F1-Score of 0.58, 0.57, and 0.49 respectively.