Our
Progress

So far we have worked on multiple feature extraction techniques to be able to obtain the ABC characteristics of melanoma (see Goals section for more information on the ABC's). We currently have been find and extract the border of a melanoma and obtain a histogram of color information of dermoscopic image.

We are currently running into some challenges with obtaining correct border values when the image contains the edge of a body part, at that point our program picks the largest border as that edge instead of the melanomas edge. Additionally, we are running into challenges with detecting the melanoma in nonuniform lighting and on individuals with more pigmented skin. To mitigate these problems we are looking into finding total variation of an image, which would look at differences in color values of the image to help detect melanoma on darker, more pigmented skin.

Grayscale + Gaussian Filter

Here we grayscale the image and use a gaussian filter to blur the image to reduce detail and smooth the edges. Then we binarize the image and overlay it with the grayscale image to see how well and how much of the melanoma we were able to detect.

Edge Detection (Method 1)

For border detection, we have two different methods to search for the edge/boundary of the potential melanoma. The first method is the edge() function which uses the sobel method, on a binarized and Gaussian blurred image of the melanoma. Next, we use the bwboundaires() function, which gives an array of the largest boundaries found in the image. The step by step process is shown in the image below.

Edge Detection (Method 2)

The second method we are looking into is the histogram of gradients (HOG) on the edge detected image to obtain border values for the melanoma. To implement this, we use the abstractHOGFeatures() function to obtain a slope field map of the 2D image to detect the edges of the melanoma.

Image Courtesy of: Kabir Deol

Detecting Color Changes of Melanoma

Following the ABC methods of melanoma detection, color changes are a key characteristic of melanoma. Therefore, to detect color changes we are taking the original RGB image and using the imhist() function to differentiate the changes in color. As show below we plotted the histogram for the individual RGB and total RGB matrices.