Chris Pollett > Students > Venkatesh

    Print View



    [CS 297 Proposal]

    [Deliverable 1 - Android App for Inter-activity Communication]

    [Deliverable 2 - Solver Performance on Custom Images]

    [Deliverable 3 - Paikin Tal Placer]

    [Deliverable 4 - Neural Networks for puzzle reassembly]

    [CS297 Report - PDF]

    [Mixed bag placer review]

Project Blog

Meeting 12: November 28, 2017

Discuss the results comparing the algorithm on regular color images versus a document. Discuss how segmentation can be improved to prevent all the chunks going into a single segment. There needs to be a tangible stitching point. Discuss about segmentation for documents when there is not much difference between segments.

Meeting 11: November 14, 2017

Discuss the possibility of using neural networks or OCR for piece similarity metrics. Professor suggested researching more on byzantine curves as a possible piece similarity metric

Meeting 10: November 7, 2017

Demo Tal Paikin Placer and show the challenges in finding the starter pieces for documents because of the usage of color gradients prediction in the current scheme

Meeting 9: October 30, 2017

Meeting cancelled as Dr. Pollett is out of town for a conference.

Meeting 8: October 24, 2017

Show a working demo of inter-piece distance calculation using current metrics on 2 different images of varying nature (B/w versus color).

To be discussed
  • The current way of calculating the average border pixel information where the channel information does not get preserved
  • Discuss about fixing a standard shred width considering the current A4 size paper standard.
  • Discuss about the n-gram model for border information
  • Ask questions about neighbor piece rotation from the excel document in the current solver
  • Discuss about a possible bug in openCV Mat data dump for color images and individual pixel values

Meeting 7: October 10, 2017

Discussed about Image dimensions and piece sizes being a factor in determining the speed of the solver. Explore the metrics currently involved in the piece border calculation and inter-piece distance calculations. Show a working demo of current metric for inter-piece distance calculation of 2 different images (A black and white one and a color image) during Meeting 8.

Meeting 6: October 3, 2017

Review and present existing literature on shredded document reconstruction. Focus on piece similarity metrics. An idea was discussed where n-grams can be used to calculate the similarity value of a shred and compare with the other shreds. The one with the highest match is the best buddy neighbor.

Meeting 5: September 26, 2017

Demo existing mixed bag solver on custom images. The challenges and issues with speed were discussed during this meeting. Specifically, the Achilles's heel of the existing solver for shredded documents reconstruction was discussed. This involved the time taken to segment pieces that had very close similarity measures and the accuracy of the assembly.

Meeting 4: September 19, 2017

Read the work on "A Fully Automated Solver For Multiple Square Jigsaw Puzzles Using Hierarchical Clustering" by Zayd Hammoudeh.
Also, begin working on making existing code for mixed bag solver work on custom images. Reading the paper mentioned above will help understand the existing code for mixed bag solver.

Meeting 3: September 12, 2017

Presented a literature review on "
Solving Multiple Square Jigsaw Puzzles with Missing Pieces" by Paikin and Tal. This is the current standard in solving mixed bag puzzles.

Meeting 2: September 5, 2017
  • Working demo of android app for inter-activity communication.
  • Clicking on a button splits the image into chunks.

Meeting 1: August 29, 2017
  • Finalized meeting time to be between 9:45 AM and 10:30 AM
  • Work on an android app to take a picture from camera and pass the image to an activity.
  • Goal of this app should be to learn inter-activity communication in android.