Understanding of face expressions plays vital role in communication as 58% part of communication contains face expressions and body language. Humans can easily identify the face expressions but it’s a very difficult task for computers. If computers will be able to identify the expressions correct then it will promote human–computer interaction. From the literature survey of research it is observed that good recognition accuracy is achieved for posed images in controlled scenario. The proposed algorithm gives good accuracy for real-time images in uncontrolled scenario and it is subject independent. Pre-processing is applied to remove noise, normalize all the images in dataset, cropping, and resizing the images. To reduce feature mismatch Gaussian normalization function is applied. To decrease intensity variation and to preserve edge information median filters are used. All the images are reshaped to 256X256 by using Bessel’s down sampling function. To capture local intensity variations, shape, edge, and texture pattern in image HOG features are calculated. These features are robust to illumination and contrast variation. For training and classification deep neural network is used. It can be trained on big dataset and it is applicable to real-time dataset. Confusion matrix is used to find other parameters like Sensitivity (recall), Specificity, FAR, FRR, Precision, and F1-score of classification model. The proposed algorithm gives satisfactory results when applied on four dataset, viz., JAFFE, CK + , FER2013, and own.