Small target detection using morphology and modified Gaussian distance function

Abstract

We propose a new small target detection system that detects small target candidates based on morphology operations and detects actual targets using a modified Gaussian distance function. To reduce clutter on the edges of clouds, a median filter is applied as preprocessing. Two kinds of images are calculated with closing and opening morphological operators, respectively. In the morphology operations, various sizes of structure elements that are used to consider the sizes of targets and candidate targets are extracted from different images between the two images in the closing and opening operations. With a modified Gaussian distance function, small targets are detected from the candidate targets. The proposed method is less sensitive to clutters than existing methods and has a detection rate of 98%. Copyright © 2014 John Wiley & Sons, Ltd.

RT-SPeeDet: Real-Time IP–CNN-Based Small Pit Defect Detection for Automatic Film Manufacturing Inspection

“…To minimize the processing time during the candidate region extraction stage, this approach adopts morphology operations that have been used in real-time small object detection studies [25,26]. Morphology operations can be used to suppress noise and enhance the features of object regions to detect small objects that appear as tiny dots in infrared images captured at a distance.…”

An Improved GMM-MRF Model for Maritime Infrared Target Detection

“…红外目标检测在许多领域中是不可缺少的关键 技术，经常应用于海上监控，船舶检测和海上救援等 等 [1][2] 。 在复杂的海洋背景下， 海水本身运动的复杂性， 以及常常伴随着较大的动态背景干扰等。此外，在红 外图像中，由于红外图像成像距离比较远，通常导致 目标成像面积少、特征不显著等特点 [3][4] 。使得红外图 像的海上目标检测变得非常地艰难。 目前，海上目标检测有以下几种：光流法、背景 减法、核密度估计法等等。光流法因海上背景杂波会 对光流场的计算结果产生影响等一系列因素，并且光 流场计算相对而言较复杂，实时性比较差，需要特别 配置的硬件支持；背景减法是一种常规的运动目标检 测方法，适用于比较单一的场景中，故在本文中所提 到的复杂场景中不能使用 [5][6][7][8] ； 核密度估计法是一种无 参数的背景建模方法，由于红外图像成像距离比较 远，通常导致这些目标成像面积少、目标特征不显著 等特点。使核密度估计法无法准确地检测海上目标图 像，另外根据核密度的原理可知其运算量较大，也很 耗时。 对于海上杂波混合高斯模型能够很快速地随场 景的变化而变化，也能够及时地检测出运动目标 [9][10] 。 文献 [9]…”