Human Action Recognition (HAR) has many challenges in research domains such as visual surveillance system, from video database extract data of content-based query and retrieval, Human Computer Interaction (HCI). Human Motion Recognition (HMR) is the research that identifies the human actions that are performed by the person body movement and is also the latest stage of the human motion capture process. This work focuses on analyzing human actions against static backgrounds. The proposed approach extracts specific features from images sourced from the Weizmann dataset to infer human actions. Experimental findings demonstrate that incorporating silhouette images significantly enhances HAR accuracy. This study employs methods such as Laplacian of Gaussian (LoG), Features from Accelerated Segment Test (FAST), Maximally Stable Extremal Regions (MSER), and Speeded-Up Robust Feature (SURF) for image feature extraction to recognize human actions. This paper aims to provide insights into their performance and practical applicability to HAR using video data. Each of the aforementioned methods brings unique strengths and capabilities, addressing aspects such as speed, robustness, and scale invariance in image feature extraction tasks.