Video Summarization (VS), is one of the recent attractive research studies in huge amount of video contents news, sports videos, TV reality shows, Movie reviews etc., Based on user’s facial expression, VS is the challenging task in digital video technology. An automatic recognition of facial expressions is the necessary task in VS used in behavioural science and in clinical practice. The reliable recognition of facial expression by a machine is the critical compared to by a human. The non-uniqueness and the deformations in the human face require the toleration in human action recognition system. This paper proposes an effective toleration of facial variations themselves and addresses the problems in facial expression and video summarization and focuses the Multi-user Facial Expressions for Video Summarization (MFE-VS). The utilization of quantitative measures in proposed work estimates the user’s viewing interest based on the human emotional perspective. Initially, Gaussian filter eliminates the unnecessary things in an input image. Then, Viola-Jones face detector detects the face and eye regions. The utilization of Histogram of Gradients (HoG) and the Local Land scape Portrait (LLP) for extraction of shape and texture features. Once the labelling of features is over, the Probabilistic Neural Network (PNN) classifier predicts the exact expression of the user in the frames. Based on the obtained expression, the proposed work extracts the interesting frames and summarizes them that constitute final stage. The variations of facial expressions of viewers constitute attention states and the classification of positive and neutral states constitutes emotional states. The comparative analysis of proposed method with the traditional methods on the parameters of success rate, precision, recall, and interest scores assures the suitability of MFE-VS in facial expression based video summarization.