Accurate tumor segmentation from CT scans of liver is a crucial stage in diagnosis. We have proposed a novel framework for automatic segmentation of tumor using Simple Linear Iterative Clustering (SLIC) technique. This approach generates super pixels and thus reduces number of regions in the segmentation. Reduced number of regions will minimize the complexity of further processing steps. The noise in the image has to be minimal for the better accuracy. For this purpose we have used median filtering as a part of the pre-processing before going for super pixel generation. Preprocessing includes noise removal and image filtering steps with resizing the images. Gray-level co-occurrence matrix (GLCM) and Histogram features are utilized for components estimation which helps for the collection of feature vectors. Finally Hamming Distance is used for validating whether a particular region is tumor or not. The experiments on various images have been carried out and results are discussed.