Exploring online ad images using a clustering approach

Abstract

Online advertising is a huge, rapidly growing advertising market in today's world. The common form of online advertising is using image ads. A decision is made (often in real time) every time when a user sees an ad, and the advertiser is eager to determine the best ad to display. Consequently, many algorithms have been developed in order to calculate the optimal ad in order to show that the current user is available at the present time. Typically, these algorithms focus on variations of the ad, optimizing among different properties such as background color, image size, or set of images but none of them define the property of objects. Our study looks at new qualities of ads that can be determined before an ad is shown (rather than online optimization) and defines which ad image's objects are most likely to be successful. We present a set of algorithms that utilize machine learning to investigate online advertising and to construct object detection models which can foresee objects that are likely to be in successive ad image. The focus of results is to get high success rate in ad image with objects appear in it. In this paper we discuss two approaches, using cascading trainer and R-CNN network. We have compare this two approaches using HOG and CNN features. R-CNN gives better result than cascading but require more time to train.