Captcha is widely used in modern websites in order to hinder illicit tasks. It provides a mechanism through which a robot and human can be differentiated. It is akin to a security system which aims at deterring fraud and automatic attack. Advanced technologies such as deep learning is capable of recognising captcha without human intervention. Needed to walk through how this security breach can happen. In addition to complex object detection and segmentation tasks, our work analysed how encoder and decoder models can be used in this task. Also,experimented with various encoders and decoders to contrast their efficiencies and to propose the best model. Typically, a convolutional neural network is an encoder and sequential model such as recurrent neural network acts as decoder. Our work proved that a combination of Resnet and Long short term memory has yielded higher accuracy. In the case of object detection approach, YOLO (achieved 0.710 mean average precision) performed better than Faster RCNN is observed. Grey scale images have also been considered as captcha recognition task is concerned with extraction of shapes rather than colours.