|Project Name||Consolidating and Adding the Image Processing Functions to LIBICV|
|Project Sudent||Mohit Daga|
|Phone Number||+91 9783582684|
This page will contain weekly targets and Updates about the work done.
Pre Coding Period
From 1st to 5th June
Will be unavailable on the internet, Visiting the meditation facility.
- Targets during this time
- As per the proposal document I will be studying the Image Processing tools.
- Read Details regarding SVN.
5th June to 9th June
- Read Codes of the utilities from Group 1-8. Made Rough Notes for all theses utilities and working for an Implementation Plan.
10th June to 12th June
- Api Design and Posting them on BRLCAD website. It can be accessed from here.
12th June - 14th June
- Working on conversion tools. Posted a patch for conversion of pix-bw and bw-pix.Patch can be found [].
Week 1 (17-22 June)
- Structure Definitions : This week structure and api definitions were designed. had good discussions with the mentors on the possibility of the design and finalizing it.
- Later during the end of week I read about high definition images. Familiarized myself with Openexr  image containers. Also about how bmp, png handles them. About
- Depth Converters
- Bw-pix, pix-bw : I converted these two utilities to functions. These were important because these can help convert a grey level image to a colour image and vice versa. Saved as patch.
- Wrote Doxygen comments for ICV Header file. Committed by ``Erik
- Revisited my pre-application ICV_LOAD patches. Improved them and posted. Waiting to be reviewed. (Patch 197).
- Cropping Functions
- Although Groups start from the second week. During Discussions with the mentors, I implemented functionalities of Group 1. icv_rect (cropping with rectangular box.) and icv_crop(cropping with skewed coordinates). Although these have to be re modified with the new structure definitions.
- I found a code redundancy while reading the utilities last week. Thus posted it. The mentors found issues with the compatibility of bwhisteq and its correctness. I checked and tested it on artificially uniform grad images, two color images with half dark and half light and natural images of lena, barbara. Turns out that this is fine.
P.S. From next week onwards I will update on daily basis.
Worked on Operations(GROUP 2). Learnt the macro usage of vmath.h. This provides nice macro definitions. Although these cannot be used for image data. Designed new macro definitions for icv_data. Initiated a discussion for keeping them in icv_math.h
- Tested three different structure definitions for the time. These are here   
- checked the Crop, and rect as per the new struct defintion
- Find out a plan to implement these structure on the existing use of a libicv in rt,libged and rmrt
26th and 27th June
- Wrote Functions to read, write, create Image and zero image.
- Separate files have been made for each format. Added pix, bw format.
- Links to these files are as follows
- Road ahead:- Incorporate them in rt/view.c rt/viewxray.c rt/viewedge.c and rt/do.c and libged/screengrab.c
- Converted the existing use of LIBICV in rt.
- Added gamma_corr field in image for gamma correction when an image's data is converted from double to unsigned char type.
- didnt do any coding work today.
- worked towards clarifying issues with my patches 171, 175, 176, 178,188,192,197,198, 201.
- did some system Cleaning and Installed new operating System and packages.
- Did code regarding Filter Functions (Group 3) and use of Kernels.
- Improved the kernel functions. tried building a test function in the src/util folder.
- Instead of going further with Designing new functions. reiterated the use of icv functions in libged/screengrab.c Changed there functions as per the new writepixel functions. Tested load, save functions of icv library.
- Wrote testing function and programs similar to libbu. These will help in testing the functions of icv. Also changed current usage of libicv in rt, rmrt and libged.
- Wrote Color_space changers. This will be useful in saving RGB data in bw format and gray data in pix format.
- Wrote crop and rect functions for icv library. Crop accepts skewed parameters and can extract image in any quadrilateral, this then converts the image to rectangular using the nearest neighbour. Whereas rect extracts rectangular part of an image.
The following milestones only include the completed code. Kernels and operations have been started but are yet to completed, thus not included in the milestone.
- Discussed with mentors about the image structures and API designs.
- Wrote save functions in new format. Implemented the following functions as per the new formats.
- Initiated a testing infrastructure for libicv. Wrote testing functions for loading and saving images.
- converted the existing use of ICV in rt,libged and rmrt.
- Implemented functions to crop image These include
- Implemented functions to change Color_space
This link takes you to the new ICV infrastructure written http://bzflag.bz/~mohit/libicv_7thJUL/
Monday (8th July)
Implemented testing routines in libicv/tests for color_space(rgb2gray and gray2rgb), pixel writing(single pixel, and pixel line). Corrected few errors in the routines. Tested color space changers with natural image(lena of 512X512 in png format). Converted that to pix(png2pix). Loaded pix image using implemented function. and converted it to gray scale image, then saved to bw format. Used this bw image to create a pix image using colorspace change to rgb. P.S. All these things are done by loading the image in double format data
Tuesday (9th July)
Implemented and tested splitting of color channels. Given a rgb image this function can split the any color channel and produce a gray scale image.
Wednesday (10th July)
- Implemented Pixsaturate as icv_saturate.
- This implementation is a function which takes in the sat value and icv_image struct. This takes the double data.
- Created a test routine for this implementation. Corrected few errors which crept in.
- Compared the results with the original routine (pixsaturate)
Thursday (11th July)
Started working on filter Again. This is a very tricky implementation as a function. There are three routines pixfilter, bwfilter and pix3filtter. Read there code along. Found some commonalities.
Friday (12th June)
Implemented the filter routine for single channel images. This is a very rough draft and will involve changing and updating the function.
Saturday (13th June)
Took a break : Invited friends for Iftikhar party
Only Tested the bwhisteq function for Method2. And updated status about previous patches.
Monday (15th July)
Improved the filter function with added generalization to dimension of kernel and number of channels.
Tuesday (16th July)
Finalized filter function. Tested it with natural images for lenna, mandrill and barbara. This implementation is highly generalized and can handle any number of channels. Also kernel dimension has been worked with (Ensuring it accepts any size of kernel). Although this file will need a relook for getting kernel coefficients of arbitrary size.
Wednesday (17th July)
Implemented icv_hist function. Also tested this file. This implementation has also been generalized for any number of channels. thus an improvement over previous code. Also it was tricky implementing this because the image data is converted to double. To deal with this used the bins concept and converted to integer.
Thursday (18th July)
Looked at plans for implementation of histeq for double data. By the look of it looks very tricky to implement this for double data. I believe I have to write it afresh.
Friday (19th July)
Implemented histeq for single channel images. Also looked at stats functions. Looked at Erik's suggestion for single functions like icv_image_sum, icv_image_mean etc.
Saturday (20th July)
Implemented stats function. Planning to merge hist functions in stats.c
Monday (22 July)
Implemented bilinear interpolation as part of scale function. Next I have to implement nearest neighbour function.
Tuesday (23 July)
Completed scale function. Now it is a generalized implementation. It is implemented as icv_scale accepts both gray scale(single channel) and rgb images. Also it works with double data.
Wednesday (24 July)
Implemented shrink function. This handles both pixshrink and bwshrink and the api function looks as icv_shrink.
Thursday (25 July)
Tested icv_scale and icv_shrink. Corrected the anomalies.
Friday (26 July)
Committed the new icv structure. Incrementally Committing all the functions and api's. Also working on a bug that crept in rt during commit.
Saturday (27 July)
Tried Fixing issue related to rtedge.
Sunday (28 July)
Monday (29 July)
Fixed few bugs which crept in during icv implementation in rt. Compiled the src code on bz account. Ran several tests on multi processor machine (bz server). Did benchmark testing. Went through large part of code in rt. Also got acquainted to multi threaded process in brl-cad.
Tuesday (30 July)
Tested and committed rect, crop, filter, filter3, fade api functions. My local tree was sort of mismanaged.
Wednesday (31 July)
Was planning to work on merging my tree in the source code, But Apparently Benchmark Error showed up. :-( Wrote icv_writeline such that now it doesn't allocate memory. This helps in putting in place the semaphores again. Even Putting semaphore didn't help resolve the benchmark issue. I took help of fellow gsoc students to get errors from benchmark, And it turned out there was issue with fractions in double data. Did some internal tests. And corrected the code.
Thursday (1 Aug)
Segregated pix, bw from fileformat.c file. Added new flags and thus modified my previous work on operations and icv_math.h Completed source tree syncing.
Friday (2 Aug)
Revisited to scale and shrink. Corrected pixshrink and bwshrink utilities. Went out for TAX registration. Could not do much work. Will make up on sunday.
Saturday (3 Aug)
Implemented bilinear [binterep(..)] and nearest neighbour interpolation [ninterep(..)] in sync with current format. Also added BOX Average shrinking [shrink_image(..)] and under sampler [under_sample()]