Evaluate Logo Detection With Deep Learning Techniques

With the ongoing development and rising popularity of computer vision technology in recent years,object detection and recognition technology have indicated its convenience and advantages in many respects such as medical and health care,national education,urban transportation.There are also a lot of applications of object detection and recognition technology in real life.Meanwhile,Logo is a significant symbol of the corporate information and logo recognition in images such as copyright infringement detection,contextual advertising placement,automated computation of brand-related statistics on social media,augmented reality,etc.However,the traditional algorithms' performance and efficiency are far from being able to meet these applications.With the development of artificial intelligence,deep learning has become one of the major breakthroughs in the field of artificial intelligence in recent years.For the moment,the convolution neural network(CNN)method in deep learning has achieved remarkable results in image classification,face recognition,pedestrian detection and has been widely applied.Although the majority of the traditional logo detection and recognition technologies in the literature are based on a single logo,this dissertation focuses on the object detection and recognition methods to detect and recognize computers' brand logo in social networks as the point of entry based on deep learning.This thesis primarily covers two aspects:The first aspect is study reviewed and examined the existing public logo recognition and recognition datasets and found that some of the databases do not fulfill our research needs and specific requirements using the manual annotation tool on our FlickrTechLogos-5 database to do the processing required,which comprises of five categories of computer brand logos.We used a web crawler to download images suitable for our research.The second aspect concentrates on three types of research and comparison algorithms and performs related experiments that have been used in recent years.We use the Caffe and Tensorflow,which are among the deep learning frameworks for network training,tuning optimization,the accuracy of recognition improvement and achieving a better training model.Finally,this thesis summarizes the different application scenarios for the three detection and recognition algorithms based on the FlickrTechLogos-5 dataset.Faster R-CNN demonstrated better performance and supremacy in the logo detection and recognition with more balanced results.