Human Facial Expression (FE) is one of the most potent and identifying or recognising FE is a difficult task. In general, a facial expression helps people to express their feelings such as sad, anger, contempt, happy, fear, disgust and surprise. In this paper the main phases of FE techniques are pre-processing, feature extraction, and classification. The different techniques involved in FE recognition and their main contributions are explained in this paper. Initially, Contrast Limited Adaptive Histogram Equalization (CLAHE) is applied to the facial expression dataset to enhance the image quality. Adaptive bilateral filtering effectively remove noise and sharpen the face image data. Scale Invariant Feature Transform (SIFT), extract features and key points for further analysis. Multi-Stream Inception-V3 is proposed to enhance the classification of facial expression images, and benefits to overcome challenges in face analysis. Using Python software the proposed Multi-Stream Inception-V3 achieved a higher accuracy of 97% which is highly efficient compared to the existing method.