Object Detection Based On Global Features

With the rapid development of electronic information technology, more and more problems related to image processing technology come into our production and life, which greatly promote the development and maturation of image processing technology. At present, the declining prices of computer equipments, the expanding internet communication bandwidth, the larger network popularity, all create favorable conditions for the development of image processing technology. As an important part of object processing, the image detection technology has received a significant development and improvement due to the efforts of domestic and foreign researchers.However, nowadays, there are still some problems left, such as poor detecting robustness, low accuracy in the object detection.Traditional object detection method, regardless of the sliding window detection, or detection of interest points around the region, is only based on local characteristics of images. However, sometimes these local features seem to be insufficient in the practical detection. First, in the case of small target object is concerned, or the features extracted from the image are of no use for the. detection, the local features become useless. Second, there are actually two closely related tasks in the images detections, namely object detection and object localization. Object detection means detecting the existence of one or multiple instances of a category of objects inside a image, which is very useful for image restoration based on objects. Object localization means finding the location and size of a specific object inside an image. Sometimes, the local probe can correctly detect the location of objects, while the object size is incorrect. But detection using global features can get the right results.This paper first gives the analysis of uncertainty in the traditional image recognition object detection. In the object detection, a modified detection method using the global features is designed and implemented. As the extra data source, this paper defined the global features as "Gist" of the scene. And then, an object detection experiment is carried out based on an object detection system. According to the experiment result analysis, the method brought out by this paper is more efficient than the traditional ones. Besides, since the computation of the gist is much easier than most of local features, this method is also faster than those traditional ones.