Nipple detection to identify negative content on digital images

In the era of the 2000s, the development of internet technology started to grow rapidly resulting in immense number of information available on the internet. However, the trend towards free internet access causes some adverse impacts on the society due to the negative contents (pornography) which are bounded on some finding information. The aim of this research is to develop a filtering system of negative content based on nipple detection, as one of the body vital parts. The proposed scheme uses Haar-Cascade classifier which is trained by 1000 positive image data (nipple images) and 8500 negative image data (no nipple-images). Beforehand, face detection is conducted to decrease misclassifying (false positive) detection around the face area. Feature extraction process uses 84 attributes of GLCM and 12 attributes of colour statistics on the nipple of the object candidates. Furthermore, MLP is conducted to classify these candidates with 10 neurons and a hidden layer for the MLP architecture. As a result, by using 160 nipple data, 12 features of colour statistics achieve the best performance with the accuracy of 90% and sensitivity of 96.3% compared to 84 GLCM features and 96 all features. In comparison to the conventional method which used 27 images data, the accuracy and specificity values are increased to 87% and 94%, respectively. However, from the consumer side, they prefer to choose the best specificity value by using 84 attributes of GLCM with specificity value of 87.34%. The consumers may be disturbed by the situation in which the non-porn images are classified as prohibited images resulting in the blocking of the system.