Object Localization In Computer Vision.

Abstract

In this work, we have introduced the computer vision discipline that includes several tasks, such as object recognition, segmentation and detection. solving the object localization task usually needs a set of features such as HAAR cascade or HOG these features must change according to the type of the localized object and this create an additional overhead. With the arrival of deep learning era it will be possible to both learn the representation subproblem (feature extraction) and the regression/classification subproblem. We leverage the concepts offered by deep learning and use them to tackle our localization problem which consists of recognizing the category of the main object (in our case cat or dog), in addition to the drawing of a bounding box around the detected object. our model mainly consists of a feature extractor termed Xception and a set of dense layers that capture the categories and the bounding box. We also notice that the proposed model has achieved a classification rate of 100% and an IoU rate of 78% on the validation dataset. As a perspective of this work, we can compare our results with more elaborated feature extractors such as SeNet,ResNet; Additionally, we can extend the presented model to tackle the detection task which assume that the image contains several objects of different classes