1. Detection of face skin cancer using deep convoluted neural network.
- Author
-
Natarajan, Gayathri, Badhusha, S. Shaik Mohamed, Monisha, R., Rajalakshmi, T., and Snekhalatha, U.
- Subjects
- *
SKIN cancer , *RADIAL basis functions , *FEATURE extraction , *IMAGE processing , *GLOBAL warming , *FACE - Abstract
With rising global warming there is an increase in incidence of cancers, particularly skin cancer. Skin cancer is often confused with moles and other skin tags and hence is detected at a very advanced stage later on. Particularly cancers on the face are misdiagnosed as acne or pimples. Hence there is a requirement for a model that particularly targets the face and identifies begin or malignant tumors in the region. Image processing has gained traction over the past few years and utilizes complex techniques that break an image down to its minute components. If can easily detect any anomaly and if combined with neural networks can also be used as a predictive model. The model developed aims to detect whether a particular image of a face shows cancerous growth or not. Python a programming Interface will be used for the purpose. Fuzzy C means classification is used for segmentation as it shows higher efficiency than K means and can segment each pixel with high accuracy. Further machine learning models such as combination of features are used to increase efficiency. Specialized texture feature extraction methods are explored in the model such as Local binary pattern coupled with GLCM and ABCD parameters. Lesions on the face will be closely studied and any images with obvious abnormalities will be studied. Supervised learning Neural network algorithm with non-knowledge based classifier will be used to predict whether the lesion in the image is cancerous or not. BPN model with radial basis function as activation function is adopted. An image set with 3000 images will be used to train the model with additional 2000 images for testing. The accuracy was observed between 89-99%. With high discriminatory power and low computational complexity, the model we have designed is assured to give outputs with low percentage of error and with maximal speed. [ABSTRACT FROM AUTHOR]
- Published
- 2022
- Full Text
- View/download PDF