CDADITagger: An Approach Towards Content Based Annotations Driven Image Tagging Integrating Hybrid Semantics

Considering the rapid growth of multimedia data, especially images, image tagging is considered the most efficient way to organize or retrieve images. The significance of image tagging is growing extensively but the frameworks employed for tagging these images aren't sophisticated. These images aren't properly tagged because of a lack of resources for tagging or manual tagging is a challenging task considering such voluminous data. Already existing frameworks take both the image data and tag-related textual data but ultimately resulted in mediocre or unpalatable performance as they are dataset centered. To overcome these limitations in existing frameworks we proposed an image tagging mechanism, CDADITagger capable of automatically tagging images efficiently and much more reliable compared to existing frameworks. This framework can tackle real-world applications like tagging a new unknown image as the framework isn't powered by dataset alone but is designed to inculcate images from search engines like Google, Bing, etc. to have comprehensive knowledge of real-time data. These images are classified using CNN and tag-related textual data is classified using decision trees for enhanced performance. While tagging images from the classified tags, are sorted based on the semantic computation values, only the top 50% of the instances classified are selected. The tags which are more correlated to the image are ranked and finalized. The proposed semantically inclined framework CDADITagger outshined the well-established frameworks with an accuracy of 96.60% and a precision of 95.84% making it a more reliable approach.