In the past few decades there has been a tremendous rise in the diet disorders, which are all attributed to unhealthy food habits. The systems include dietary assessment systems that can take pictures of meal in and analyze that meal for nutritional balance and it is very handy for the improvement of their bad eating habits and help one have a healthy life. To this extent, this research study presents a new approach to this problem that consists of making the system to estimate food attributes like nutritional values with requiring an actual input of the image of the food. In our approach, we use several deep learning models for the task of food classification. Attributes are also computed and estimated out from the large amount of semantically similar words from text data collected through the World Wide Web and collections. When we carried out the equivalent of experiments and used a dataset of 100 classes with an average of 1000 images per class, we obtained a maximum top 1 classification rate of about 85%. Sub-continental foods are also added to an extension of the benchmark dataset called Food-101. As the quantitative analysis demonstrates, the efficiency of the proposed system is intrinsic to the nature of the Food-101 dataset and, by extension, to the sub-continental foods list.