A strategy of determining the degree of body exposure in images Jang Seok-Woo1,*, Byun Siwoo2 1Assistnat Professor Department of Software, Anyang University, 22, 37-Beongil, Samdeok-Ro, Manan-Gu, Anyang, 430–714, Republic of Korea 2Professor, Department of Software, Anyang University, 22, 37-Beongil, Samdeok-Ro, Manan-Gu, Anyang, 430–714, Republic of Korea *Corresponding Author: Siwoo Byun Professor, Department of software, Anyang University, Korea Email: swbyun@anyang.ac.kr
Online published on 16 October, 2018. Abstract Background/Objectives In this paper, we suggest a novel method to accurately obtain the human umbilical region, which is an element representing the noxiousness of the images, by using representative features and robust learning algorithms.. Methods/Statistical Analysis In the suggested method, we first detect the nipple region of the human using the color feature and the brightness feature from various types of input images, and detect the candidate areas of the navel using the positional information relative to the detected nipple area. Here, we define and apply a nipple map using color and brightness features. Then, only the actual navel regions among the detected candidate areas are detected through filtering using a learning algorithm.. Findings The experimental results of this study show that the suggested harmful image detection algorithm extracted human navels more accurately than the traditional detection method from various types of input color images. In order for the comparative evaluation of the accuracy of navel region detection algorithms, the method of the nipple detection and the position relation from the detected nipples in the proposed method, and the whole proposed method using AdaBoost algorithm were evaluated. In the conventional method, the region adjacent to the straight line that vertically and equally divides the line connecting the centers of the already detected two nipple regions is selected as a candidate region. The proposed method based on extended Haar-like features and AdaBoost algorithm more accurately detected a navel region. In other words, the method using simple positional relation with nipples failed to accurately detect candidate navel regions and caused wrong detection, whereas the proposed method was able to accurately detect candidate regions of the navel through the AdaBoost learning and recognition algorithm.. Improvements/Applications The suggested umbilicus detection approach is expected to be used effectively in various real applications related to the detection of harmful image contents. Top Keywords Feature extraction, Input image, Geometric information, Morphological operator, Color. Top |