Research On Advertisement Logo Detecting Approach In Videos

With the accelerated pace of commercialization in the field of modern medium, videomedia advertising elements are growing rapidly in videos, and higher requirements arealso asked in video supervising works. As the monitor of medium markets, videosupervisors often need to detect the time position and the length of existing time of anadvertisement logo which appears in the video, in order to standardize the commercialbehavior of the advertisers. Therefore, the relevant aspects of the technology has beenpopular in today’s medium field, and also been the essential assurance of the healthydevelopment of media industry.Since some related technologies involve emerging fields such as image processing,image understanding, lack methods and theoretical basis, the researches of thesetechnologies are also considerably difficult. For example, there are always variances suchas size, rotation and light with the advertisement logos displayed in the videos, it isunable to adapt to a variety of variances by using pixel-based comparison orcontent-based image detecting approaches. In addition, as an image sequence in the timeaxis, the additional time dimension also adds up to the difficulty of the video detectingwork. The lack of methods and theories hindered researches in related fields, and alsoobstructed the development of relevant industries. As the theory basis of this thesis, theprocedures and principles of SURF algorithm is briefly introduced firstly; this thesis putsforward an image detecting approach combining corners detection and SURF descriptors, makes some improvements to make it adapted to the special features of advertisementlogos and videos, forms a set of video detecting approach, and then designs a videodetecting software, and testifies it in practical work. So, according to this, the thesiscontains following content:1. Starting from the matching and comparison between the images, and taking size,rotation, perspective, noise variations and other factors into account which the imagesmay have, the procedures and principles of SURF algorithm which is adaptive and robustto these variations are introduced.2. Considering about the speed of approaches, this thesis puts forward an imagedetecting approach combining corners detection and SURF descriptors based on SURFalgorithm. After a brief introduction of various types of corner detecting method, cornersdetection and SURF descriptors calculating are combined to form new improvedapproaches. After solving the problem of applicability to scale-variance of images whileretaining the rapidity, the image detecting approaches are testified by using actualadvertisement logos, and compared with traditional SURF algorithm to analyze thesuperiority of the approach.3. Since the image detecting approach is still not good enough in the practicalapplication of the advertisement logo matching, according to the special features ofadvertisement logos, a various of improvements such as matching ratio and multi-templated image matching are made, in order to make it more suitable for advertisementlogo detection.4. After making some improvement for the image detecting approach, the approachwill be applied to the actual video detection. This thesis describes the structure of thevideo at first, and makes some improvements according to the special features of videos,and the method of using this approach in videos is described, in order to accomplishadvertisement logo detection successfully, and the feasibility and effectivity of theapproach is analyzed. 5. Finally, according to the theory of software engineering, this thesis designs a videodetection software by using the method based on data flow, and through thedemonstration of the software operation, verifies the effectivity of the software with anexample, in order to eventually reach the destination of detecting advertisement logosaccurately and quickly in videos.