Re: [help-GIFT] [help-gfit]About the number of the features

[risc]

On Tue, May 08, 2007 at 08:03:25PM +0800, Mingchung Liu wrote:
> In some article, you said that , The occurence of a given colour in a
> particular block is treated as a binary feature. For 256*256 images there
> are thus 56440 possible color block features.
> 
> But in your code(see below) ,you never treated it that way .
> You just calculate 340 block features, of which each block has only one.
> 
> Am I right? If I am wrong ,please tell he where did you add them into the
> array named  featrue[feature_index]?
> 
> ===========================================
>  /***** count the features *****/
> 
>    /* we know that there will be a mode color for each block */
>    num_features = num_total_colour_blocks;
> 
>    /* and there will be one for each non-zero histogram entry. Convert
>    histogram entries to range [0, FREQ_MAX] while we're at it */
>    for (i = 0; i < colmap_size; i++) {
>        col_histogram[i] =
> (int)(rint(FREQ_MAX*(double)col_histogram[i]/(double)(square(image_size))));
>        if (col_histogram[i] != 0)
>            num_features++;
>    }
> _______________________________________________
> help-GIFT mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/help-gift

no, your reading of the code is right, we use 340 blocks. thats :
256 (16*16) 16 pixel wide blocks, 
64 (8*8) 32 pixel wide blocks, 
16 (4*4) 64 pixel wide blocks, and
4 (2*2) 128 pixel wide blocks.

i of course, havent read the article you're refering to, just re-wrote that 
code block. blame me. :)

Julia Longtin <address@hidden>