Transition detection is the necessary step in retrieval and investigation of videos on the basis of contents. Since last two decades, most of the researchers are involved in developing algorithms for detection of gradual transitions. However, the features of all gradual transitions are different and hence this issue still needs to be addressed precisely. After identifying this issue, an integrated shot boundary detection method is proposed in this paper. Wipe transition is the effect which is mostly used in the video making industries .Since there are different types of wipe transitions, it becomes very difficult to detect such transitions. Due to complexity in detection of wipes due to noise, object and camera motion earlier methods have less focus on this issue. In this paper an efficient wipe detection method is presented which gives better results even in the presence of object and camera movements.
Shot boundary detection is the first step in the video processing applications. The detection of gradual transition is quite difficult compared to the detection of abrupt transition. Among the different gradual transitions, the detection of wipes is more challenging because the wipes are of different categories and having complex nature. We have presented a novel method for wipe transitions detection. In the proposed algorithm we used Normalized Mean of Approximate Wavelet Co-efficient (NMAWC) as a metric for potential wipe transition detection. To apply algorithm on whole video to detect wipes is computationally expensive. So we tested our algorithm on the video clips containing wipe transitions. The experimental results are evaluated using the performance metrics Recall, Precision, F1 measure, detection rate and the results are compared with the existing methods. Our proposed algorithm achieved a relatively better trade-off between recall and precision as compared to other algorithms. We tested the proposed approach on different wipe effects including Special wipes. The proposed method successfully avoids false positives caused due to object/camera motion and illumination.
Thinning is one of the most important preprocessing steps in the character recognition. But this process has certain limitations like low speed and deformation. To eliminate this problem, skeletonization is used, where the character to be recognized is skeletonized. This paper describes how characters are recognized by skeletonization algorithm which is trained by neural network. Here for better understanding and experimentation, we are considering categories of decorative characters. Here, we are using an algorithm based on neural network, which determines the representative points and connections making up the skeleton by combining AVGSOM non-supervised learning. The proposed method has been applied in images with different characters and their rotations along with scaling. The results obtained are compared to existing stored database, showing quite encouraging results with more than 90% recognition efficiency. Finally, some conclusions, together with some future scopes are presented.
The detection of dissolve transition is more difficult than detecting fade in and fade out. In this transition, the last frames in the previous shot fade out and the beginning frames in the next shot fade in i.e. the overlapping of fade out and fade in occurs. The dissolves may be of three or more number of frames. In some videos very short dissolve of even three frames also occurs. So, there are lots of challenges in detection of dissolves. Also illumination and camera / object motion gives rise to false positives thereby degrading the algorithm performance. Many researchers addressed this issue but could not achieve the robustness in the presence of illumination and object / camera motion. Therefore this issue needs to be resolved. An algorithm has been proposed for dissolve detection. In this algorithm, color histogram difference between consecutive frames is calculated and average value of this difference for all consecutive frames is used as a metric for dissolve detection.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.