SVP

Back

Academic Year 2004-2005

March 17, 2005

March 16, 2005

TODO:
1. Prepare for Mel's presentation. Slides, maybe?
2. Reflection Questions.
3. Thesis.
        I am thinking: PDF --> Intepretor --> Diagram Parser.
    Question: Why not using SVP?
                        What's the research point?
4. Documentation of PJ/DJ

Feb.3, 2005
worked on PJ/PJX to extract vector diagram from BMC pdf articles.
BMC pdf articles at /proj/bkl/myshao/BioMedCentral/cdrom
PJ: /proj/futrelle/pdf/pj-1.10, /proj/futrelle/pdf/myshao/workTMP/PJ
PJX: /proj/bkl/myshao/pjx-1.3.5, /proj/bkl/myshao/pjx-1.2.1

The following note is from an email to Bob.

I am running PJ and DJ code on the BMC PDF files, to see if the code has become rusty. I tested on nine PDF files, two of which gave us good results, i.e., some vector digrams. The failure of the others lies in two kinds of exceptions, one is about PDF color operator SCN/scn, the other is from some unrecognized tokens like '<0050'.

Both of these two exceptions come from a PJ class called StreamParser.

The first exception is about ICCBased color spaces. I once got stuck with it when I worked on pdf two years ago, and noticed that it's a bug. I modified the related code so that it worked at that time.

Before I fix the bug by myself, I think maybe PJX has worked on it. However they didn't. I will deal with the bug.

To the second exception, I have no clear idea yet.

BTW, the color operators SCN/scn can be found on Page 199 in the PDF reference (/proj/futrelle/pdf/PDFRef1.3.pdf). The StreamParser.java is
at /proj/bkl/myshao/doc/TAR/pjx-1.2.1/src/com/etymon/pj/StreamParser.java, and if you go to line 627, you will notice there is a note from the
author.

Feb. 1 2005
BlogCMS looks a good one, both blog and forum.
However, when I created an account, I notice it works not as I expected.
Need to know more about PHP so that I can understand and modify it.

working on the coop evaluation forms, finished half of students' forms. Next, merge the students' information with the rubric matrix.

Jan. 20, 2005
1. one-wing extension
    One-wing extension is critical in the intersection detection. In the case of the intersection between two lines of similar orientations, the four endpoints are so far away from each other that neither two-wing nor core extension can push the endpoints close enough to meet. The one-wing extension that extends without touching any other line segments can be applied to push the endpoints into the intersection as far as possible to meet other endpoints. In another case of bar graph, the bars usually vertically meet the x-axis, which make the detection of x-axis difficult because the inner side of x-axis is not clear (due to the intersections with bars). The one-wing extension, instead, is able to makes use of the outer side of x-axis which is clear to detect it.
2. Orientation adjustment
    The orientation of a line segment was decided by the orientation of the first PCA. Generally we may use the average of the chained PCAs' orientation as the segment's orientation. However, we need to think carefully about how to adjust the orientation when chaining the adjacent PCAs together. Two extremely cases may be curves and two touching lines with slightly different orientations.
3. Connected components
    Supposing a scan box has only one connected component (CC), we can define a PCA based on that CC. However sometimes a scan box may contain more than one connected components. We need to examine each of these CCs in the box, and decide which one is the major CC. We then accept the major CC and define its PCA for the scan box. All the other minor CCs are rejected. A major CC is a CC whose weighted center is roughly the center of the scan box.
4. Occlusion
     Partial matching information can be used in SVP to detect objects overlapping with each other. We need to explore what other people did on topics like 'partial matching', 'partial feature/signature', etc.

Jan 2005
HOW ABOUT create a weblog for myself?

Jan 18,2005
Drop ECEG315 class. I am not ready for it yet, since it requires a lot of signal processing stuff.
More needed to do, both research and portfolio.

Research:
Pdf, PJX Etymon
Papers, google, citeseer, web of science
connected component, tick mark, occulution. Code
books: handbook for scholars, prob., safari of OReilly

COOP:
php connector for weblog, forum, and e-portfolio
Excel survey file
XML template Excel for rubrics matrix

Nov 2004
On Forum and Weblog, found some free software. Need to find a good one.

Oct 2004

Oct 18 (good progress today)
        Discussed with Bob about how to evaluate the vector level performance. Bob wants to just compare the endpoints to see how near they are to each other. I don't think so. Because each groundtruth line may broke into several lines in the detected results. I planned how to evaluate them.
        Create a pair of groundtruth and detected diagrams. 7005.jpg and 7005-gt.svg.
        Need to parse the svg file to get the groundtruth vectors(some code online). Using JDOM to parse SVG xml file. Using regular expression (more)to extract the path operators.

        Read some article about XSLT and XML. for portfolio. good progress. Need more.


Oct 17 2004
        pixel level evaluation is done by StatisticsInfo.java.
        Now focus on vector level. GroundtruthLine.java, GroundtruthDiagram.java, Evaluation.java. Transform svg file to groundtruth java objects, then use Evaluation.java to compare
        them with the detected lines(in data file, need unserialize).
        GREC'95 and GREC2003 have some test images and groundtruth. /proj/bkl/myshao/SVP/GREC95Contest and /proj/bkl/myshao/paper/GREC2003/
        Found they use CAD drawing as groundtruth and the line attributes can be known in advance, this is a difference between their work and ours. ALso a challenge for us.

HURRY UP!

September 2004

Sep. 27th
        coop group meeting, and made plan of what I need to do before next meeting which include:
better understanding of osp2.0, find some document of it. like how to customize.....
journal/weblog, how to connect it with osp. read two documents Mel gave me. and prepare to talk with him next Monday.
how to let students to write journal during their coops? Is there any open source systems we can use?
try blogger myself.

Bob emailed me about work progress report. Need a plan for every week. (BEFORE THE END OF OCTOBER)
create 2 or 3 groundtruth vector images.
use SVP to get best results.
implement basic evaluation, like statistical information.

Sep.17th
        health insurance issue
        11:00am
       read paper, wrote summary at /proj/bkl/myshao/paper/Summary/papersummaries/summaryevaluations.tex
        1:00pm


Sep.16th
        10:40am
        1 hour on OSPI, still no progress
        turn to research
        iMac still can't connect to www, so the work of jpeg->svg has to stop currently. Waiting our system guys to deal with it.
        still has some confusions about ground truth vector graph.
        How did other people get groundtruth?
        SVG specification, and svg browsers. download Batik 1.5 at /proj/bkl/myshao/doc/batik-1.5.1/. TO run it, java -jar batik-squiggle.jar.
        Performance evaluation papers may give me some idea.(GREC97's contest report)
        find a software which can convert raster images to vectors(www.softelec.com), unfortunately it can only be intalled on Windows systems.
        Similarily, some other famous raster to vector software like r2v(www.ablesw.com/r2v/, www.r2v.com/,www.algolab.com/r2v.htm) also require Windows. And, they mostly work on CAD and GIS.
        Now, get a list of raster to vector conversion companies:

Easy Trace PRO	www.easytrace.com
R2V for Windows	www.ablesw.com/r2v/index.html
TracTrix 2000	www.trixsystems.com/tractrix.html
IRAS/B	www.intergraph.com/imaging/raster_prods.asp
GTXRaster R2V	www.gtx.com/products/detail.asp?id=3
Image Series	www.hsgt.com
VP HybridCAD	www.hybridcad.com
Draftsman Products	www.arborimage.com/aihome.htm
Vectory	www.graphikon.com/english/vectory.html
UCLID Software	www.uclid.com

        1:00pm
        Now, go to read papers and books.

Sep.15th
        9:30am
        Coop Meeting (First meeting this semester)
        build OSPI 1.5
        several problems, solved some
        read through OSPI forum, got some idea

Started: Jan. 12, 2004

Week5 Week4 Week3 Week2 Week1

Week4 Summary
Feb. 3 Tuesday
            11:00am -2:00pm and 3:40pm - 5:30pm add skills to portfolio
            6:00pm - 9:00pm CSG131

            need to do: prepare for code walk
                                install phpmyadmin
                                paper reading, book reading
                                review CSG131
Feb 2, Monday
            10:30am - 2:00pm SVP debug, experiments
            3:00pm - 5:00pm meet with Bob. (Disc, thesis, code walking on Wed)
            need to do: properties, print code, read books/papers
                                customization(skills, outlook)

Week3 Summary
Jan. 30 Friday

Jan.29, Thursday

Jan.28, Wednesday
            paper report

Jan.27, Tuesday
            customization to have coop in ospi
Jan.26, Monday
            Group meeting
            meet with Bob

Week 2 Summary

    Jan.23 Friday
                11:00am -12:00pm Write summaries.
                12:30pm - 2:30pm Meet with Bob, about papers and SVP parameters. Line Width should be considered in svp as changable parameter.
                3:30pm - 4:40pm write up summaries
    Jan.22 Thursday
                11:00am -12:00pm OSPI
                12:00pm - 2:00pm Met with Mario about OSPI and php
                2:00pm -6:00pm        Read two papers about moment analysis

    Jan. 21 Wednesday
                10:30am - 2:00pm OSPI, can modify the elment. but has some problem with upload path.
                2:00pm - 3:00pm Met with Viera and Mario
                   3:30pm Went back home, New Year's Eve
    Jan. 20 Tuesday
                10:30am - 5:00pm OSPI, try to modify an element, AcadPlan. but got some problem
    Jan.19 Monday
                10:00am - 4:00pm     SVP parameters and one paper

Week Summary
                Once I started to work, I worked pretty good and effeciently, like Monday and Wednesday. It seems like I need some pression to make me start. This is not good. I should always push myself. To do like that, I need a plan like what to do in this week, in this month, in this semester.
                Please do come to office everyday, except extremely bad weather. Only in this way, can I arrange my time wisely.
                Should begin to work sooner, like 10:00am, or 10:30am.
                Should use the nighttime more efficiently.
                Have a nice weekend and better next week!

Jan.16 Friday
                11:00pm - 1:30pm SVP reading, parameters, packages, documentation
                1:30pm - 2:30pm AI semianr first meeting. Next meeting maybe two weeks later.
                2:30pm - 4:00pm SVP

Jan.15 Thursday
                Got sick, stay at home

Jan.14, Wed
                11:00am - 12:00am OSPI
                12:00 - 1:30pm COOP reflection meeting
                2:00pm - 2:30pm Meeting with Viera
                3:00pm - 5:00pm Meeting with Bob

Jan. 13: Tuesday
                11:20am - 1:30pm OSPI find dataobject, action, model, etc
                1:30pm - 2:30pm search papers of research. maybe what the topic is not that important, the thing is do some research: read, write, think.
                2:30pm - 4:00pm OSPI. not effective. Need rest.

       enn, I've spent enough time for ospi, 20hrs this week.

Jan. 12, Monday
               10:30am - 2:00pm work on portfolio
               2:00pm - 3:00pm coop meeting
                3:00pm - 5:00pm meetingwith Bob. plan on survey/research paper, on SVP, image processing, software dev. PLAN
                5:00pm - 6:00pm portfolio
                9:00pm - 11:00pm work on portfolio
                12:00pm - 2:00pm OSPI/php