$1\mathrm{.}1$ Objective

The objective of this assignment is to design an algorithm that segments lesions from healthy skin

regions automatically. For developing the segmentation algorithm, you are given a set of $80$

dermoscopic images and their corresponding ground truth. Figure $1$ shows lesion image samples and

segmentations.

Figure $1$: Skin lesion image $($top$)$ and its ground truth segmentation $($bottom$).($Visualization: White

represents lesion area, and black represent healthy skin ares $($background$))$

$1\mathrm{.}2$ Guidelines

$$ Partial collaboration policy applies. You can lightly discuss these questions with your course

mates when you are working on the assignment but you need to declare with whom you had

discussed with as responsible collaborators. Include a statement at the end of the report to indicate

the nature of your collaboration.

$$ In regards to the objective of this assignment, you are NOT allowed to use third$-$party packages to

assist you in this problem $($including deep learning models$).$ Packages 'pip'$-$ed from Python

Package Index $($PyPI$)$ are acceptable.

$$ You can use Spyder $($which comes with iPython console$),$ but you can use your own favourite IDE

or opt for the basic text editor and command line.

TDS$3651$ VISUAL INFORMATION PROCESSING TRI. $2310$

$2$ Scripts and Functions

$2\mathrm{.}1$ Codes to Write

Your working function $`$segmentImage$\text{'}$ that you need to write is contained within

'imageSegment.py$\text{'}$ def

segmentImage $($img$)$: # write

your code here $...$ return

outputImg

The inputs and output of the $`$ segmentImage $\text{'}$ function are as specified follows: inputImg

: Input image, a $3$D numpy array of row$*$col$*3$ in BGR format outputImg : A $2$D numpy array

segmentation mask where the lesions are represented with the

following intensity values:

Segment Background Lesion

Intensity $01$

No visualization codes or functions are provided. You can write your own in a separate script for

purpose of visualizing the outputs or to generate nice figure$/$plots for reporting.

$2\mathrm{.}2$ Evaluation Functions

An evaluation function is provided to test your algorithm:

evaluateSegment.py:

Evaluate the Adapted Rand Error$[1]$

$,$ pixel wise precision and recall, and the Intersection over Union

$($IoU$)$ between a set of output segmentation with the corresponding ground truth segmentation. It also

returns the average evaluations for the image set.

$[1]$ Arganda$-$Carreras, Ignacio, et al$.$ "Crowdsourcing the creation of image segmentation algorithms for connectomics."

Frontiers in neuroanatomy $9(2015)$: $142.$

The functions are runnable on Anaconda Prompt or standard command$-$line prompt $($if necessary path

settings have been configured$).$ You can use the $`-$h$\text{'}$ switch to get further help on how to use these

functions, and what other options are there.

NOTE: You are NOT ALLOWED to change the code of this function except changing the Default

Parameters, which include the image directories, number of images in the directory, and

verbose to select the evaluation output.

TDS$3651$ VISUAL INFORMATION PROCESSING TRI. $2310$

$2\mathrm{.}2\mathrm{.}1$ Package Requirement

The vanilla Anaconda installation does not come with the PrettyTable package. Please install

via pip at Anaconda Prompt.

$>>$pip install prettytable

$2\mathrm{.}2\mathrm{.}2$ Example of Usages

These commands shows evaluation of all images and specific lesions respectively:

$>>$python evaluateSegment.py $-$v

$>>$python evaluateSegment.py $-$p

This command segments and evaluates the whole image set in the directories specified under the