CMPT 487: Computer Vision and Image Processing
计算机视觉和图像处理代考 This is a take-home exam. Here’s what this means for the purposes of this exam:The exam is an individual effort.
Final Examination (Take-Home)
Due April 29, 2021, 6:00pm, Absolutely No Extensions
This exam consists of 9 pages containing a total of 11 questions worth a total of 66 marks.
Instructions 计算机视觉和图像处理代考
Exam Rules
This is a take-home exam. Here’s what this means for the purposes of this exam:
The exam is an individual effort. Submissions must be entirely your own work (see heading Aca- demic Honesty, on page 2).
Between the time the exam is released and the submission deadline, you may not consult/use any other resources (including Internet resource) or tools except those mentioned below, nor may you speak to anyone not mentioned below on any matter related to the exam (whether they are in the class or not).计算机视觉和图像处理代考
You may consult any official CMPT 487 class materials, such as the lecture slides, lecture videos, exercises, exercise solutions, assignments, and assignment solutions.
- You may use a calculator, you mayuse a dictionary.
- Youmay ask questions of the course instructor (see heading Asking Questions, below).
For the duration of this take-home exam you should not share access to your computer with anyone else writing the same take-home exam if it contains files with exam CMPT 487/819 exam answers that are accessible to other users.
Submitting Your Exam 计算机视觉和图像处理代考
The completed exam is to be submitted to Canvas. The submission page is in the same place that assign- ment submissions are found. Follow the Assignments link on the left menu, and click on the CMPT 487 Final Exam assignment (do *not* use the CMPT 819 Final Exam link). Submit your exam in the same way you have submitted your previous programming assignments.
Multiple submissions are permitted until the deadline passes, but only the most recent submission will be graded. Download your submissions after uploading to double check it contains every- thing needed. Leave yourself time to do this!
- Youare solely responsible for ensuring that all required files are correctly
- At the end of this document is a Hand-in Checklistthat tells you exactly what to submit.
- Do not wait until the last minute to submit! It is your responsibility to give yourself time to submit and verify your submission.
Asking Questions 计算机视觉和图像处理代考
Do not post questions about the exam in any public place. This includes Canvas discussions and Discord.
Treat the asking of questions as if you are in an invigilated exam room and can only privately ask an instructor.
This is final exam. We will answer questions only about the wording of a question, or clarification about what a question is asking you to do. We will not tell you whether an approach or an answer is correct or incorrect or even on the right track. This is no different from an invigilated, in-person exam.
- Ifyou have a question send a single email addressed to the course instructor:
– eramian@cs.usask.ca
Your question will be answered as soon as possible. You should not expect immediate answers, so leave yourself time to ask the questions.计算机视觉和图像处理代考
- The instructor will answer questions from the time the exam is released until 5pm on Thursday, April 29 (about 1 hour before the deadline). Start early, so you can ask questions early. Do not expect replies to emails timestamped after 5pm, Friday, April 23.
Academic Honesty
This exam is an individual undertaking – cheating on an exam is considered serious academic miscon- duct by the university and can be met with disciplinary action, including suspension or expulsion. By submitting your exam solutions to Canvas, you are affirming that the submitted work is entirely your own.
Breaching the above rules of the take-home exam is academic misconduct. Any evidence of academic
misconduct on this exam will be pursued to the fullest extent possible through formal academic mis- conduct hearings. This means that the Office of the Dean will be informed immediately, no informal resolution option will be given, and you will need to appear before the hearing board. You do not want to be before the hearing board.计算机视觉和图像处理代考
- (6points) Consider the following images and their histograms:
Each of the images A, B, and C pictured above, has been computed from the original 205 232 pixel image using a grayscale transformation. The transformations used (in no particular order) were:
- Histogramequalization
- Negation
- Gamma transform with γ <0
For each of images A, B, and C, identify which transformation produced the image, and explain what characteristics of each histogram you used to deduce the transformation used (i.e. how do you know from the histogram what the transformation is?).计算机视觉和图像处理代考
2.Recall that histogram equalization is performed on grayscale images by using the cumulative density function.
(a)(2points) If the histogram p(r) (expressed as probabilities) for an input image with intensities r
between 0 and 8 is:
r 0 1 2 3 4 5 6 7 8
p(r) .02 .08 .10 .40 .1 .05 .05 .1 .1
then compute the transformed intensity for pixels of intensity 4 that would result from histogram equalization. Show all of your work.计算机视觉和图像处理代考
(b)(1 point) Why does the classical histogram equalization algorithm not always produce a uniform (flat) histogram for digitalimages?
- Consider the following 5 by 5 pixel greyscaleimage I:
10 5 60 20 5
10 5 60 20 5
10 5 60 20 5
10 5 60 20 5
10 5 60 20 5
Assuming zero-padding of the image, determine the result of filtering the middle row of I with
(a)(5 points) a 3 × 3 medianfilter;
(b)(5 points) a filter with mask
In both cases, show all your calculations.
4.(5 points) Consider the followingimages:
Image A is the original floating-point grayscale image, image B is the binary image resulting from thresholding image A with a threshold of 0.41. The remaining binary images, C, D, E, and F are the result of binary morphological operations performed on the thresholded image B.
(a)(4 points) For each of images C, D, E, F, identify the morphological operation (dilation, erosion, opening,closing) used on image B to obtain Justify your answer — describe the characteristics of the image that you used to deduce the answer, that is, explain how you know which operation was used.计算机视觉和图像处理代考
(b)(2 points) The same structuring element was used to produce images C, D, E, and What was itsshape? Justify your answer — how do you know (how did you deduce your answer)?
5.(4 points) Consider a dataset where you want to isolate bright objects on a dark background.
The objects generally have a width and height of several tens of pixels and the width and height of the imagesare on the order of several hundreds of Under perfect imaging conditions and assuming 8-bit grayscale images, the background would show up as intensity 25, and the foreground objects would show up as intensity 175. Suppose, however, that in practice, the imaging conditions are far from ideal and that the images tend to contain Gaussian noise with a standard deviation of 32 intensity levels. Further suppose that you have been tasked with using a bilateral filter to de-noise the images prior to further processing.计算机视觉和图像处理代考
Describe how you would design the bilateral filter. Provide all details and parameters of the design of the spatial and spectral components of your bilateral filter, and justify your choices.
- (5points) Suppose that you have implemented a fully automatic segmentation algorithm for a dataset where there is one object of interest per image. The dataset consists of 3000 images, and their cor- responding ground truth segmentations. Of these images, you selected 100 images which you used repeatedly during the development and refinement of your segmentation algorithm. Now you need to perform a more comprehensive evaluation of your final
Design an evaluation methodology for your newly implemented segmentation algorithm using the data that you have available. Describe the methodology in detail. Justify your choices.计算机视觉和图像处理代考
- (5points) For the given image function f , determine the non-symmetric (non-normalized) co-occurrence matrix M(1,1) (also called M45◦,1). That is, for the direction: \.
8.(5 points) Suppose you are creating a fully automatic image classification algorithm that is designed to label input images with one object of interest per image.
You have a dataset of 8,500 images available and you know the correct class label for each image. Since you don’t have very much data, you have elected for a traditional machine learning approach (i.e. not deep learning), and you have determined thatthere are three possible classifiers that might be viable.
Design a classifier evaluation methodology to assess the performance of the classifiers on your dataset and determine which would be the best one to use. Describe the methodology in detail. Justify your choices. You may assume that there are equal consequences for any wrong prediction.计算机视觉和图像处理代考
- (2 points) Below are two ROC curves obtained for two different classifiers for solving the same com- puter assisted diagnosis problem for cancer detection. Based on the ROC curves, justify which classi- fier is better for this
10.(5 points) Compare and contrast traditional machine learning for computer vision and deep learning with convolutional neural networks. Discuss fundamental similarities, differences, as well as advan- tages/disadvantages of each.
11.(10points) Suppose you are asked to design an image processing/computer vision pipeline for license plate detection and The system expected to operate on images acquired from inexpensive dashboard cameras which you can assume have a maximum resolution of 640 by 480 pixels. The system must be able to locate the license plate region in an image of a car taken from directly behind.计算机视觉和图像处理代考
For the current stage of development it will be sufficient if the system can identify the Canadian province of the plate, and need not be able to actually read the characters on the license plate. Using the knowledge you have gained from CMPT 487, describe how you would design a processing pipeline for such a system. What specific steps would you use? How will you locate the plate region within the image? How will you determine its province? Assuming you had as many labeled images as you wanted, what evaluation methodologies and metrics would you use to assess the performance of the various stages in your pipeline? Justify your design decisions.计算机视觉和图像处理代考
Your answer should not exceed two handwritten pages or one typed page (10 point font, 1in margin). Example images of Canadian license plates are on the following page.
Examples of Canadian provincial license plates.
Hand-in Checklist 计算机视觉和图像处理代考
When you submit your exam, choose one of the following options:
Q Submit a single document containing all of your answers to all exam questions, with question numbers clearly labeled, and answers must appear in numerical question order. Acceptable file formats are: Word (.doc or .docx), PDF (.pdf). Your document may include embedded photographs of handwritten answers or diagrams accompanying typed answers provided they are clearly legible. You may not submit separate images files if you choose this option; images must be embedded with in the DOC or PDF file.
Any other document file formats not listed above be considered “no submission” and will not be graded.
If you choose this option, you may just hand in your single document file as is, you do not need to put it in a ZIP file.
OR
Q Submit a ZIP file archive of images that are photographs of your handwritten answers. The file names of the images should be page1, page2, page3 (with appropriate extensions) and so on so that it is absolutely clear which order the images are meant to be viewed in. Answers be clearly labeled with their question number, and answers must appear in numerical question order in the series of numbered images. Handwriting in images must be clearly legible. Acceptable image formats are: JPEG (.jpg or .jpeg), PNG (.png) or TIF (.tif).
Remember that ZIP files are the only acceptable archive format. Submissions received in 7zip, RAR, or other non-ZIP archive formats will be considered “no submission”, and will not be graded.
其他代写：program代写 cs作业代写 app代写 Programming代写 homework代写 考试助攻 finance代写 代写CS finance代写 java代写 course代写 代写CS作业 考试助攻
