Image and video classifications, are important problems in multimedia content understand that requires bridging the gap between the target semantic categories, or classes, and the low-level visual descriptors that can be automatically obtained. At the same time, they are valuable tools towards other applications like object detection and recognition, visual content description, semantic metadata generation, indexing and retrieval.This work aims to segment the video into a number of shots using three different types of algorithms, classify the video frames data into static and dynamic blocks depending on the difference between the blocks of successive frames, extract two types of features from the static and dynamic blocks of the video shots, the adopted features are:
- The statistical features (mean, standard deviation, mean absolute deviation, skewness).
- The textural features (energy of gradient, contrast, modification fractal dimension H).
The discrimination power for the extracted features was determined by using each features alone, or combinations of features (up to 5. features). The test results indicated that some combinations of these features are useful to successfully recognize the video shots from each other especially when the extracted video sequences from any shot consist of video frames more than (25) frame the results also showed that the features extracted from the static blocks (except the H) gave higher discrimination power than the features extracted form the dynamic blocks.Finally the K-means clustering algorithm was used to categorize the video shots into a number of classes. The test results indicate that this algorithm shows good stability in classifying the video shots, because most of the extracted sequences from each shot were classified as members to the same class.
