Identifying and correcting imbalanced labelled image for multi-label image annotation

Eternally increasing number of image documents accessible on the social media unremittingly stimulates research of improved and accessible annotation models and additional well-organized retrieval techniques which custom mash-up of available data on scenes, objects, events, context and emotion. It is very smart to develop imbalanced labelled image dataset for multi-label annotation applications. The image classification has become an incredibly active research topic, with widespread applications. Most methods for visual recognition are fully supervised, as they make use of bounding boxes or pixel wise segmentations to locate objects of interest. It permits us to train K Nearest Neighbor (KNN) with images from social networks where images are only weakly labeled with several labels are incorrectly labeled. This algorithm prolongs and extends the k-NN classifier to multi-label data. In the proposed system, we decided to detect and rectify the imbalanced labelled images by employing an algorithm called KNN (K Nearest Neighbor). In KNN an object is categorized by a widely held vote of its neighbors, with the object being dispensed to the class utmost common among its k nearest neighbors. Optical character recognition is also called as optical character reader which is an Electronic transformation of images of inputted, handwritten or in black and white printed text into machine-encoded text, whether accessible from a skimmed and scanned document or from description text superimposed on an image. In the proposed system, OCR is used to spot on the inappropriate text into accurate text. The main motivation of our design is to satisfy both the multi-label scenario as well as rectifying the incorrect labels in the images. Experimental results on real web images demonstrate the effectiveness of the proposed method.