Human identity is the most important feature representation that is used in most of the real world applications. There are various algorithms that take the credit for face recognition, however there is very little research focused on identification of humans when face recognition fails that includes when the data for facial recognition is partially available. However very limited knowledge is available whether person recognition can be performed based on the back pose alone. To deal with the above said concern, we had proposed an algorithm using pix2pix Gan earlier in our work. Now in this paper we are proposing an optimized approach to overcome the concerns we had in our earlier paper. This proposed algorithm deals with optimizing the UNET layer which is used as a key component in the generator phase of pix2pixGAN allowing us to increase the accuracy and efficiency of the results. We have also focused on advocating the concept of color normalizing to get better results in attaining human recognition using the back pose data, i.e., when an individual is seen from behind. Our finding shows that back pose based human identification is challenging but can be performed using deeplabv3+. Semantic segmentation enhances the processing recognition when seeing the person from behind.