Dictionary Learning Features Based Cross Media Retrieval Using Minkowski Distance
Alugoju Sreelatha, C.V. Gopala Krishna Rao, Porika Sammulal · International journal of intelligent engineering and systems · 2018
Cross Media Retrieval (CMR) is one of the emerging research areas in the field of internet services and multimedia technology.The primary objective of this paper is to develop an effective dictionary learning methodology to reduce the semantic gap between the low-level and high-level features.In this experimental research, the image and text data in the Wikipedia dataset are separated, after separating the image and text data, feature extraction is performed individually.A suitable combination of feature extraction methods (normalized histogram colour feature and bag-ofwords) are under-taken for image and text feature extraction.The normalized histogram colour feature has very quick generation, compared to other feature vectors.In addition, Bag-of-words measures the repetition of the words in the bag and retrieves the data similar to the query data.Also, it removes the conjunction words and the remaining words are named as keywords.These keywords are used as the query text and it compared to the proposed dictionary learning.Then, retrieves the text and image related to the keywords based on Minkowski distance measure.Finally, the experimental outcome shows that the proposed approach delivers better performance in terms of Mean Average Precision (MAP) value, retrieval efficiency, precision and recall.The proposed methodology improved the MAP value up to 0.24-0.20 compared to the existing methods.