.zip

包括基于SIFTGPU的sift特征提取，sift匹配，基于ransac误匹配点剔除等(SIFTGPU based sift feature extraction, sift matching, mismatching points based ransac removed, etc.)

ITTI视觉显著性检测MATLAB工具箱可用于检测图像视觉显著性(ITTI visual saliency detection MATLAB toolbox can be used to detect visual saliency of images)

图像库中有240幅完整的和残缺的飞机和舰船遥感图像。可用hu矩对其特征提取，用于分类。(Image library has 240 complete and incomplete aircraft and ships remote sensing images. Available hu moments of its feature extraction for classification.)

说明：  该课题为基于MATLAB的汽车出入库识别系统，带有丰富的人机交互GUI界面。目前毕业设计选题中，传统的中规中矩的车牌识别不易得到高分，必须要在此基础上有所创新方得可以避开其他雷同课题，不会轻易被导师被否决而导致毕设不过。因此建议在车牌识别基础上加入多种方法的对比，看看哪种方法的精度高。虽然目前有很多方法的车牌识别，但是都是在各自的测试库里面测试的，比如测试的车牌图像不同，字符库不同，导致无法做对比。整个设计在一个GUI界面上完成。(This project is based on MATLAB vehicle in and out of the warehouse identification system, with rich human-computer interaction GUI interface. At present, in the topic selection of graduation project, the traditional regular license plate recognition is not easy to get high scores, so it is necessary to innovate on this basis to avoid other similar topics, and it will not be rejected easily by the tutor, leading to the failure of graduation project. Therefore, it is suggested to add a variety of methods on the basis of license plate recognition to see which method has high accuracy. Although there are many methods of license plate recognition, they are tested in their own test library, for example, the test license plate image is different, the character library is different, so it can not be compared. The whole design is completed on a GUI interface.)

为了提高数字图像中椭圆检测的效率和准确性,提出了一个基于最小二乘法的改进的随机椭圆检测算法. 该算法随机选取图像中的3 个边缘点,在以这3 个点为中心的窗口内,从边缘点中拟合出可能椭圆,并通过随机选 取的第4 个边缘点来确认可能椭圆. 利用直接最小二乘法椭圆拟合的特性,引入可能椭圆边缘点收集和椭圆重新 拟合的迭代过程来提取最终的椭圆参数. 通过对含有不同噪声的仿真图片和包括残缺椭圆的实际图片的实验表 明,新算法的改进是有效的. 与原算法相比,新算法降低了对参数的依赖性,提高了检测的速度、稳定性和准确性, 同时保留了原算法的抗噪声能力.(: An imp roved randomized ellip se detection algorit hm based on least square approach was proposed to enhance t he efficiency and accuracy of ellip se detection in digital images. This algorit hm randomly se2 lect s three edge point s in t he image , and t hen uses least square approach to fit all t he edge point s in three windows , which are defined by t he t hree edge point s. The forth edge point is randomly selected to judge whet her a possible ellip se exist s in t he image. Utilizing the characteristic of direct least square fit ting of ellip se , an iteration process of edge point collecting and ellip se refit ting of possible ellip se was int roduced to ext ract t he final ellip se’s parameters. Artificial images wit h different level s of noise and nature images containing incomplete ellip ses were employed to test t his algorithm. Experimental result s show that t he improvement s are notable. Compared with the original algorit hm , t he propo sed algorit hm reduces)

用于SAR图像的相干斑抑制，内有增强Lee滤波，增强Kuan滤波，增强GammaMap滤波算法，还有等效视数，边缘保持指数的求法(For speckle suppression of SAR images, there are enhanced Lee filtering, enhanced Kuan filtering, enhanced GammaMap filtering algorithm, and the calculation of ENL and EPI.)

猫群算法在图像分类中的应用。猫群算法是一种比较新的算法，目前研究和使用的并不多。(The cats swarm optimization in image classification. Cat swarm optimization is a relatively new algorithm, not much research and use.)

本算法实现了dct基字典的构造并用其进行对图像的消噪。(Algorithm DCT-based dictionary structure and use it for image de-noising.)

附带GUI界面，可用于各种图像的特征点提取，并对两幅图像进行配准拼接，使用的是sift算法( U9644 u5E26G0 u754C u972 u53EF u7528 u4E8E u5404 u79V U8FDB u884C u914D u51C6 u62FC u63A5 uFF0C u4F7F u7528 u7684 u662Fsift u7B97 u6CD5)

混合高斯模型使用K（基本为3到5个） 个高斯模型来表征图像中各个像素点的特征,在新一帧图像获得后更新混合高斯模型,用当前图像中的每个像素点与混合高斯模型匹配,如果成功则判定该点为背景点, 否则为前景点。通观整个高斯模型，他主要是有方差和均值两个参数决定，,对均值和方差的学习,采取不同的学习机制,将直接影响到模型的稳定性、精确性和收敛性。由于我们是对运动目标的背景提取建模，因此需要对高斯模型中方差和均值两个参数实时更新。为提高模型的学习能力,改进方法对均值和方差的更新采用不同的学习率 为提高在繁忙的场景下,大而慢的运动目标的检测效果,引入权值均值的概念,建立背景图像并实时更新,然后结合权值、权值均值和背景图像对像素点进行前景和背景的分类。(Gaussian mixture model using K (essentially 3-5) Gaussian model to characterize the features of each pixel in the image, in the image of the new frame for updated Gaussian mixture model, with each pixel in the image with a Gaussian mixture current model matching, if successful, determined that the point of the background points, otherwise the former attraction. Throughout the entire Gaussian model, he mainly has two parameters determine the variance and the mean, the mean and variance of the study, to take a different learning mechanism, will directly affect the stability, accuracy and convergence model. Since we are moving object extraction of the background modeling, so the need for the Gaussian model variance and mean two parameters real-time updates. In order to improve the learning ability of the model, an improved method for updating the mean and variance of different learning rates to improve in the busy scene, large and slow moving object detection results, the introduction of)