Generating New Melanoma Data Using a Combination of Generative Adversarial Network and Local Binary Pattern

The available melanoma skin cancer dermoscopy databases have low and unbalanced images with non-uniform illumination that make melanoma detection methods challenging. To address these problems, in this paper, we propose a new method to generate new data including melanoma. In fact, our new proposed method combines generative adversarial network and local binary pattern. On the other hand, first, the images existing in the dataset are fed to the generative adversarial network. Then, many new images are generated and finally, the local binary pattern is applied to them. Therefore, the number of the new generated data is large and balanced and the generated data does not have illumination changes. Also, these data show useful and meaningful features that increase the difference between melanoma and nevi. The experiments have shown that the proposed method has a good effect on increasing the accuracy of melanoma diagnosis. According to the results, the proposed method has increased the convolutional neural network's efficiency by 7%.