This paper proposes a novel algorithm to detect shot changes in a video stream using Autoassociative Neural Network (AANN). Histogram features are extracted from the video frame and AANN is used to capture the distribution of the features. The proposed AANN Misclustering Rate (AMR) algorithm is used to detect the shot transitions. The algorithm is evaluated using various factors and the best attributes are chosen. The experiments demonstrate the effectiveness of the proposed algorithm in detecting shots of less than two seconds duration.

Developing organizational methods and structural analysis of video is the crucial step for video retrieval and indexing systems. Unlike the growing content of digital video, availability of reclamation methods has not risen at the pace. This is due to the fact that the multimodal nature of video data makes it flabby for traditional text-based retrieval methods. For analyzing a video, the basic choice is to fragment the video into more manipulatable and executable pieces. Shot is a sequence of video frames having similar characteristics.

A shot consists of continuous frame sequences captured by a single camera action. Depending on whether the transition between shots is abrupt or gradual, the shot boundaries can be categorized into two types: abrupt transition or cut and gradual transition (GT).

From the literature, the shot transition techniques may be classified based on the features and the methods used. Feature-based techniques use color space, texture, shape information and motion vector. The STD problem and the major issues involved are described in [1] and [2]. A formal study on shot boundary detection, which presents an idea for detecting boundaries using graph partition model, is given in [3, 4]. Foveated technique for video segmentation is used in [5]. RGB histogram values are used in [6] to compare the average values of each color channel in every frame. Color histogram in RGB space is used in [7]. An inter-frame similarity measure based on motion is obtained using a block-matching process [8]. STD approaches using neural network [9] and Support Vector Machine (SVM)-based [10] methods are also discussed in the literature. [11] presents an approach for detecting MTV video shot using Hidden Markov Models (HMMs) which uses color, shape and motion features. A method which performs video segmentation via active learning is proposed in [12]. A dual method based on self-adapting dual-threshold comparison is adopted in [13] for detecting shot transitions. [14] proposes shot boundary detection in soccer video using twin-comparison.

Science and Technology Journal, Shot Transition Detection, Video Indexing, Autoassociative Neural Network, Organizational Methods, Shot Transition Techniques, Graph Partition Model, Hidden Markov Models, Evaluation Metrics, Shot Detection Algorithms, Gradual Transitions, Digital Technology, Error Measurements.