Face recognition is an important application for machine vision and machine learning methods. This assessment will guide
Question:
- Face recognition is an important application for machine vision and machine learning methods. This assessment will guide you to study how different combinations of techniques influence the performance of a simple face recognition system, and write a report about your findings.
Detailed instructions:
Get faceRec.py from NOW and run it in your Python environment. (Note: The first time you run it, it downloads a large data set of face images from the internet, which requires several hundred MBs of free hard disk space. Also, if you get an error message related to a function „urllib.urlretrieve“, this is likely caused by connectivity problems, so try again several hours later. If the problem persists, contact your tutor.)
Modify faceRec.py in a number of different ways, and write a report containing descriptions of your methods and results:
- Replace the learner in the original program with a one-layer Perceptron using scikitlearn. Create a subsection in your report describing all parameters of the perceptron and training method, and how they were set. Also report the accuracy of the resulting method.
- Replace the learner in the original program with a deep convolutional neural networkusing the keras or TensorFlow library (do not use dimensionality reduction with this learner). Create a subsection in your report describing all parameters of the neural network and training method, and how they were set. Also report the accuracy of the resulting method.
- Use methods from scikit-image to find the eyes in each picture of the data set (it maybe easiest to select a random subset of suitable size and display them together with the inferred position of the eyes as images using matplotlib. Accuracy can then be reported by counting the number of correctly marked eyes.) Create a subsection in your report describing your method of detecting the eyes, and the resulting accuracy.
- Use a clustering method of your choice to cluster the images, and measure howaccurately the clusters correspond to the sets of images of one person. (You should not use dimensionality reduction methods like PCA for this task, but you may decrease the resolution of the images if you need to speed things up). Also, investigate whether using one chosen image pre-processing method from scikit-learn (e.g., a variant of histogram equalisation, or edge detection) can improve this accuracy. Create a subsection in your report where you mention your methods and the resulting accuracy.
- Take a set of images of only two different persons. Apply PCA as in the originalprogram. Then pass a suitable number of principal components to a genetic programming system that is implemented using the deap library, and evolve a function that outputs a positive number if an image is from one person, otherwise a negative number. Measure the accuracy of the evolved classifier. Create a subsection in your report where you describe your methods and results.
The report should have a brief introduction where you summarize what you did. It is not necessary to write a literature review. However, if you took a significant portion of code from somewhere, you must reference it. You must also reference any scientific paper or other source that you used for deciding which methods or parameters to use. For this assignment, it should not normally be necessary to take significant portions of code from anywhere except the online documentation of the libraries used. You must not use any external libraries other than those taught in the module. You must not use any direct file, system or web access methods in your code. (Images can be loaded from files or the Internet using methods provided by the taught libraries.)
Both your report and your code must be submitted electronically on NOW before the deadline. The code must be submitted in TWO forms: (a) all Python files, (b) the whole code from all files as an appendix to your report.
Statistics For The Behavioral Sciences
ISBN: 9781111830991
9th Edition
Authors: Frederick J Gravetter, Larry B. Wallnau