Recognize Candidate Objects as Artificial Text in the Movie

Objectives: Text characters which appear in a video sequences regarding a valued source of important information for content-based indexing and retrieval applications. Methods/Analysis: It's known that the text characters are difficult to be extracted and recognized due to their various sizes, grayscale values and complex backgrounds. This article will introduce a new hybrid methodology based on morphology and chain code for recognizing the identified candidate objects. Findings: Applying this morphology operation (thinning, miss-hit) on the candidate objects will enable the determining the number of terminal-points and 3-connection points to construct the feature code for each object. This can be used in comparison between the feature codes for each object with a dictionary that will be created by training sample of characters. If the feature code corresponds to isolated features code in the dictionary, this will enable the system to retrieve the character name, but if the feature code matches more than one or does not match any feature code in the dictionary; it will use the second filtration operation to recognize the remaining objects by using chain code for those objects. This leads to check the chain code of the remained objects, if there is correspondence between checked chain code and the isolated chain code in the dictionary then the character name will be retrieved. Otherwise, the system can be concluded that the object is not a character. Novelty/Improvement: The experimental result shows that the proposed system has better results when compared to previous works that depended on chain code only. Thus, it can greatly reduce the time complexity, as well as, extract and recognize all cases except in the case of an overlapping between objects in the movie.