Saliency Based Automated Image Cropping
Abstract
Automatic Image Cropping have been enormously popular in the fields of photography, printing industries and in many other fields. It can be usually done in two ways; attention based and aesthetic based. Attention based approach mainly focuses on the important subject in the image whereas aesthetic based approach focuses on the attractiveness of the image. In this project, we proposed a novel attention based approach, which crops out the distracting objects from the image. In our approach, spatially weighted dissimilarity saliency model is used for object detection. In contrast to many methods, cropping of the detected object is done based on the centroid which is obtained from the weighted mean formula. The region of cropping starts from the centroid and is expanded by comparing the surrounding values of centroid with threshold value. The cropped region covers 80% of the total saliency values. Our experiment shows the accomplishment of qualitative image cropping and even the efficiency in terms of time complexity. Moreover, we can demonstrate improvements of our method over recent cropping algorithms on a broad range of images.
Collections
- OSU Theses [15752]