I am trying to write a function
[offset,coffset]=findLowNhbr(map)
that for each pixel in a map finds the eight neighbors to the pixel, and returns two matrices with both the row and column offsets to the lowest neighbor (uses the numbers -1, 0 and 1). Border pixels are given 0 offsets for both the row and column, since they do not have neighbors.
Here is what I think the general plan for this function should be:
For each point, find the eight nearest neighbors.
If the neighbor is lower than the point, return -1
If the neighbor is at the same elevation as the point, return 0
If the neighbor is higher than the point, return +1
Store these offsets in two matrices.
I am at a complete loss as to where to start, so any advice or questions are welcome!
Not entirely sure what you mean, but here's something to get you well on your way:
neighbors = cell(size(map));
for ii = 2:size(map,1)-1
for jj = 2:size(map,1)-1
% current element
M = map(ii,jj);
% extract neighbors
N = map(ii-1:ii+i, jj-1:jj+1);
% compare values and store
neighbors{ii,jj} = M<N - M>N;
end
end
This will result in a cell-array neighbors, which contains the same number of elements as map, but each entry looks something like this:
>> neighbors{2,3}
ans =
0 -1 1
1 0 -1
1 0 -1
which is the information on all neighbors of pixel (2,3).
Edit: This is how you can add inf to all sides of the map, assuming that map_original is your original map.
map=inf(size(map_original)+2)
map(2:end-1,2:end-1) = map_original
Assuming you have padded the map with infs on all sides, here is something to get you started:
area =-1:1;
for i=2:size(map,1)-1
for j = 2:size(map,2)-1
bestfound=inf;
bestk=0;
bestl=0;
for k = area
for l=area
if k~=0 && l~=0
Like i said, this will only get you started !
Related
This temporary matrix has written in Matlab, could anyone please give me some explaination?.
%Temporary matrix for evaluation
EE=E;
n_neighbour_high=1;
for i=1:x_size
for j=1:y_size
%Check neighbours
k=i-1:i;
l=j-1:j;
%If neighbours within the grid
if min(k)>0 && min(l)>0 && max(k)<=x_size && max(l)<=y_size
%If among all the neighbours there enough high elemets, all the block is high
if sum(sum(double(EE(l,k)>0)))>=n_neighbour_high
E(l,k)=elev;
end
end
end
end
let's break it down:
for i=1:x_size
for j=1:y_size
means you are looping over every element of your 2d-Matrix.
%Check neighbours
k=i-1:i;
l=j-1:j;
Here you are defining new variables called kand lwhich are basically the number before i and j up to i and j.
assuming i = 2 would make k a vector with the values [1 2]. If i would be 3 then k would be [2 3] and so on... always the number before i or j and the number i or j.
%If neighbours within the grid
if min(k)>0 && min(l)>0 && max(k)<=x_size && max(l)<=y_size
Here you are simply checking if the indices are valid (inside the range 1 and the 2d-array size) (Some of these checks are unnecessary by the way)
This is necessary to avoid the problem for the case i or j = 1 where the values in k or l would get [0 1].
%If among all the neighbours there enough high elemets, all the block is high
if sum(sum(double(EE(l,k)>0)))>=n_neighbour_high
E(l,k)=elev;
end
Here you are taking a sub 2d-Matrix from EE (EE(l,k)) of size (2x2) and you are counting how many of those 4 elements are bigger then 0 and (sum(sum(double(EE(l,k)>0)))).
E(l,k)=elev;
If the number of elements are bigger then your threshold n_neighbour_high``you are saving the value ofelevat exactly those positions into the matrixE`.
I think it would really help you to learn to use a debugger (click the line number in your matlab editor so that you get a small red dot)
have a look at this
Assuming that I have a dataset of the following size:
train = 500,000 * 960 %number of training samples (vector) each of 960 length
B_base = 1000000*960 %number of base samples (vector) each of 960 length
Query = 1000*960 %number of query samples (vector) each of 960 length
truth_nn = 1000*100
truth_nn contains the ground truth neighbors in the form of the
pre-computed k nearest neighbors and their square Euclidean distance. So, the columns of truth_nn represent the k = 100 nearest neighbors. I am finding difficult to apply nearest neighbor search in the code snippet. Can somebody please show how to apply the ground truth neighbors truth_nn in finding the mean average precision-recall?
It will be of immense help if somebody can show with any small example by creating any data matrix, query matrix, and the ground truth neighbors in the form of the pre-computed k nearest neighbors and their square Euclidean distance. I tried creating a sample database.
Assume, the base data is
B_base = [1 1; 2 2; 3 2; 4 4; 5 6];
Query data is
Query = [1 1; 2 1; 6 2];
[neighbors distances] = knnsearch(a,b,'k',2);
would find 2 nearest neighbors.
Question 1: how do I create the truth data containing the ground truth neighbors and pre-computed k nearest neighbor distances?
This is called as the mean average precision recall. I tried implementing the knearest neighbor search and the average precision recall as follows but cannot understand (unsure) how to apply the ground truth table
Question 2:
I am trying to apply k nearest neighbor search by converting first the real-valued features into binary.
I am unable to apply the concept of k-nearest neighbor search for different values of k = 10,20,50 and to check how much data has been correctly recalled using the GIST database. In the GIST truth_nn() file, when I specify truth_nn(i,1:k) for a query vector i, the function AveragePrecision throws error. So, if somebody can show using any sample ground truth that is of similar structure to that in GIST, how to properly specify k and calculate the Average precision recall, then I shall be able to apply the solution to the GIST database. As of now, this is my approach and shall be of immense help if the correct way is provided using any example that will be easier for me to relate to the GIST database. The problem is on how I can find neighbors from the ground truth and compare it with the neighbors obtained after sorting the distances?
I am also interested on how I can apply pdist2() instead of the present distance calcualtion as it takes a long time.
numQueryVectors = size(Query,1);
%Calculate distances
for i=1:numQueryVectors,
queryMatrix(i,:)
dist = sum((repmat(queryMatrix(i,:),numDataVectors,1)-B_base ).^2,2);
[sortval sortpos] = sort(dist,'ascend');
neighborIds(i,:) = sortpos(1:k);
neighborDistances(i,:) = sqrt(sortval(1:k));
end
%Sorting calculated nearest neighbor distances for k = 50
%HOW DO I SPECIFY k = 50 in the ground truth, truth_nn
for i=1:numQueryVectors
AP(i) = AveragePrecision(neighborIds(i,:),truth_nn(i,:));
end
mAP = mean(AP);
function ap = AveragePrecision(rank_id, truth_id)
truth_num = length(truth_id);
truth_pos = zeros(truth_num,1);
for j=1:50 %% for k = 50 nearest neighbors
truth_pos(j) = find(rank_id == truth_id(j));
end
truth_pos = sort(truth_pos, 'ascend');
% compute average precision as the area below the recall-precision curve
ap = 0;
delta_recall = 1/truth_num;
for j=1:truth_num
p = j/truth_pos(j);
ap = ap + p*delta_recall;
end
end
end
UPDATE : Based on solution, I tried to calculate the mean average precision using the formula given formula hereand a reference code . But, not sure if my approach is correct because the theory says that I need to rank the returned queries based on the indices. I do not understand this fully. Mean average precision is required to judge the quality of the retrieval algortihm.
precision = positives/total_data;
recal = positives /(positives+negatives);
precision = positives/total_data;
recall = positives /(positives+negatives);
truth_pos = sort(positives, 'ascend');
truth_num = length(truth_pos);
ap = 0;
delta_recall = 1/truth_num;
for j=1:truth_num
p = j/truth_pos(j);
ap = ap + p*delta_recall;
end
ap
The value of ap = infinity , value of positive = 0 and negatives = 150. This means that knnsearch() does not work at all.
I think you are doing extra work. This process is very simple in matlab, you can also operate on entire arrays. This should be faster than for loops, and is a bit easier to read.
Your truth_nn and neighbors should have the same data, if there are no errors. There is one entry per row. Matlab already sorts the kmeans result in ascending order, so the column 1 is the closest neighbor, the second closest is column 2, 3rd closest is 3,.... There is no need to sort the data again.
Just compare truth_nn to neighbors to get your statistics. This is a simple example to show you how the program should go. It will not work on your data without some modification
%in your example this is provided, I created my own
truth_nn = [1,2;
1,3;
4,3];
B_base = [1 1; 2 2; 3 2; 4 4; 5 6];
Query = [1 1; 2 1; 6 2];
%performs k means
num_clusters = 2;
[neighbors distances] = knnsearch(B_base,Query,'k',num_clusters);
%--- output---
% neighbors = [1,2;
% 1,2; notice this doesn't match truth_nn 1,3
% 4,3]
% distances = [ 0 1.4142;
% 1.0000 1.0000;
% 2.8284 3.0000];
%computes statistics, nnz counts number of nonzero elements, in the first
%case every piece of data that matches
%NOTE1: the indexing on truth_nn (:,1:num_clusters ) it says all rows
% but only use the first num_clusters columns. This should
% prevent the dimension mistmatch error you were getting
positives = nnz(neighbors == truth_nn(:,1:num_clusters )); %result = 5
negatives = nnz(neighbors ~= truth_nn(:,1:num_clusters )); %result = 1
%NOTE1: I've switched this from truth_nn to neighbors, this helps
% when you cahnge num_neghbors
total_data = numel(neighbors); %result = 6
percent_incorrect = 100*(negatives / total_data); % 16.6666
percent_correct = 100*(positives / total_data); % 93.3333
I need to assign weights to edges of a graph, from the following papers:
"Fast linear iterations for distributed averaging" by L. Xiao and S. Boyd
"Convex Optimization of Graph Laplacian Eigenvalues" by S. Boyd
I have the adjacency matrix for my graph (a 50 by 50 matrix), with 512 non-zero values.
I also have a 256 by 1 vector with the optimal weights.
For the software I'm using, I need a 50 by 50 matrix with the weight of edge (i,j) in the relevant position of the adjacency matrix (and with the opposite sign for edge (j,i)).
My attempt is below, but I can't get it working.
function weights = construct_weight_mtx(weight_list, Adj)
weights = zeros(size(Adj));
positions = find(Adj);
for i=1:length(positions)/2
if Adj(i) == 1
weights(i) = weight_list(i);
end
end
weights = weights - weights';
find(Adj) == find(weights);
end
You're finding the nonzero positions in the original adjacency matrix, but you're finding all of them. To get around this, you then take only the first half of those positions.
for i=1:length(positions)/2 ...
Unfortunately, this takes the indices from complete columns rather than just the positions below the diagonal. So if your matrix was all 1's, you'd be taking:
1 1 1 0 0 ...
1 1 1 0 0 ...
1 1 1 0 0 ...
...
instead of:
1 0 0 0 0 ...
1 1 0 0 0 ...
1 1 1 0 0 ...
...
To take the correct values, we just take the lower triangular portion of Adj and then find the nonzero positions of that:
positions = find(tril(Adj));
Now we have only the 256 positions below the diagonal and we can loop over all of the positions. Next, we need to fix the assignment in the loop:
for i=1:length(positions)
if Adj(i) == 1 %// we already know Adj(i) == 1 for all indices in positions
weights(i) = weight_list(i); %// we need to update weights(positions(i))
end
end
So this becomes:
for i=1:length(positions)
weights(positions(i)) = weight_list(i);
end
But if all we're doing is assigning 256 values to 256 positions, we can do that without a for loop:
weights(position) = weight_list;
Note that the elements of weight_list must be in the proper order with the nonzero elements of the lower-triangular portion ordered by columns.
Completed code:
function weights = construct_weight_mtx(weight_list, Adj)
weights = zeros(size(Adj));
positions = find(tril(Adj));
weights(positions) = weight_list;
weights = weights - weights.'; %// ' is complex conjugate; not a big deal here, but something to know
find(Adj) == find(weights); %// Not sure what this is meant to do; maybe an assert?
end
I am new to Matlab and am trying to implement code to perform the same function as histeq without actual use of the function. In my code the image colour I get changes drastically when it should not change that much. The average intensity in the image (ranging between 0 and 255) is 105.3196. The image is of an open source pollen particle.
Any help would be much appreciated. The sooner the better! Please could any help be simplified as my Matlab understanding is limited. Thanks.
clc;
clear all;
close all;
pollenJpg = imread ('pollen.jpg', 'jpg');
greyscalePollen = rgb2gray (pollenJpg);
histEqPollen = histeq(greyscalePollen);
averagePollen = mean2 (greyscalePollen)
sizeGreyScalePollen = size(greyscalePollen);
rowsGreyScalePollen = sizeGreyScalePollen(1,1);
columnsGreyScalePollen = sizeGreyScalePollen(1,2);
for i = (1:rowsGreyScalePollen)
for j = (1:columnsGreyScalePollen)
if (greyscalePollen(i,j) > averagePollen)
greyscalePollen(i,j) = greyscalePollen(i,j) + (0.1 * averagePollen);
if (greyscalePollen(i,j) > 255)
greyscalePollen(i,j) = 255;
end
elseif (greyscalePollen(i,j) < averagePollen)
greyscalePollen(i,j) = greyscalePollen(i,j) - (0.1 * averagePollen);
if (greyscalePollen(i,j) > 0)
greyscalePollen(i,j) = 0;
end
end
end
end
figure;
imshow (pollenJpg);
title ('Original Image');
figure;
imshow (greyscalePollen);
title ('Attempted Histogram Equalization of Image');
figure;
imshow (histEqPollen);
title ('True Histogram Equalization of Image');
To implement the equalisation algorithm described on the Wikipedia page, follow these these steps:
Decide on a binSize to group greyscale values. (This is a tweakable, the larger the bin, the less accurate the result from the ideal case, but I think it can cause problems if chosen too small on real images).
Then, calculate the probability of a pixel being a shade of grey:
pixelCount = imageWidth * imageHeight
histogram = all zero
for each pixel in image at coordinates i, j
histogram[floor(pixel / 255 / 10) + 1] += 1 / pixelCount // 1-based arrays, not 0-based
// Note a technicality here: you may need to
// write special code to handle pixels of 255,
// because they will fall in their own bin. Or instead use rounding with an offset.
The histogram in this calculation is scaled (divided by the pixel count) so that the values make sense as probabilities. You can of course factor the division out of the for loop.
Now you need to calculate the accumulative sum of this:
histogramSum = all zero // length of histogramSum must be one bigger than histogram
for i == 1 .. length(histogram)
histogramSum[i + 1] = histogramSum[i] + histogram[i]
Now you have to invert this function and this is the tricky part. The best is to not calculate an explicit inverse, but calculate it on the spot, and apply it on the image. The basic idea is to search for the pixel value in the histogramSum (find the closest index below), and then do a linear interpolation between the index and the next index.
foreach pixel in image at coordinates i, j
hIndex = findIndex(pixel, histogramSum) // You have to write findIndex, it should be simple
equilisationFactor = (pixel - histogramSum[hIndex])/(histogramSum[hIndex + 1] - histogramSum[hIndex]) * binSize
// This above is the linear interpolation step.
// Notice the technicality that you need to handle:
// histogramSum[hIndex + 1] may be out of bounds
equalisedImage[i, j] = pixel * equilisationFactor
Edit: without drilling into the maths, I can't be 100% sure, but I think that division by 0 errors are possible. These can occur if one bin is empty, so consecutive sums are equal. So you need special code to handle this case too. The best you can do is take the value for the factor as halfway between hIndex, hIndex + n, where n is the highest value for which histogramSum[hIndex + n] == histogramSum[hIndex].
And that should be it, once you have dealt with all the technicalities.
The above algorithm is slow (especially in the findIndex step). You may be able to optimize this with a special lookup datastructure. But only do that when it's working, and only if necessary.
One more thing about your Matlab code: the rows and columns are inverted. Because of the symmetry in the algorithm, the result is the same, but it can cause puzzling bugs in other algorithms, and be very confusing if you examine pixel values during debugging. In the pseudocode above I used them the same as you, though.
Relatively few (5) lines of code can do this. I used a low contrast file called 'pollen.jpg' that I found at http://commons.wikimedia.org/wiki/File%3ALepismium_lorentzianum_pollen.jpg
I read it in using your code, run all the above, then do the following:
% find out the index of pixels sorted by intensity:
[gv gi] = sort(greyscalePollen(:));
% create a table of "approximately equal" intensity values:
N = numel(gv);
newVals = repmat(0:255, [ceil(N/256) 1]);
% perform lookup:
% the pixels in sorted order need new values from "equal bins" table:
newImg = zeros(size(greyscalePollen));
newImg(gi) = newVals(1:N);
% if the size of the image doesn't divide into 256, the last bin will have
% slightly fewer pixels in it than the others
When I run this algorithm, and then create a composite of the four images (original, your attempt, my attempt, and histeq), you get the following:
I think it's convincing. The images are not exactly identical - I believe that is because the matlab histeq routine ignores all pixels with value 0. Since it is fully vectorized it is also pretty fast (although not nearly as fast as histeq by about a factor 15 on my image.
EDIT: a bit of explanation might be in order. The repmat command I use to create the newVals matrix creates a matrix that looks like this:
0 1 2 3 4 ... 255
0 1 2 3 4 ... 255
0 1 2 3 4 ... 255
...
0 1 2 3 4 ... 255
Since matlab stores matrices in "first index first" order, if you read this matrix with a single index (as I do in the line newVals(1:N)), you access first all the zeros, then all the ones, etc:
0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 ...
So - when I know the indices of the pixels in the order of their intensity (as returned by the second argument of the sort command, which I called gi), then I can easily assign the value 0 to the first N/256 pixels, the value 1 to the next N/256 etc, with the command I used:
newImg(gi) = newVals(1:N);
I hope this makes the code a little easier to understand.
I'm building a program that plays Connect 4, and one of the things to check for is cases where your opponent has the following board positions:
[0,1,1,0,0] or [0,1,0,1,0] or [0,0,1,1,0]
where your opponent is one move away from having three pieces in a row with a blank on either side. If you don't fill one of the middle elements on your next move, your opponent can go there and force a checkmate.
What I have is a board of 42 squares, numbered 1:42. And I created a matrix called FiveCheck, where each row maps to five consecutive board positions. For example:
FiveCheck(34,:) = [board(7),board(14),board(21),board(28),board(35)];
FiveCheck(35,:) = [board(14),board(21),board(28),board(35),board(42)];
are two of the diagonals of the board.
I can test for the possible checkmate with
(sum(FiveCheck(:,2:4),2) == 2 + sum(FiveCheck,2) == 2) == 2
And that gives me a vector with 1's indicating that the corresponding FiveCheck row has a possible checkmate. Let's say the 34th element of that vector has a 1, and the pattern for that diagonal (from the example given above) is [0,0,1,1,0]. How do I return 14, the board position I should move to?
Another separate example, if the 35th element of that vector has a 1, and the pattern for that diagonal is [0,1,0,1,0], how do I return 28?
EDIT: I just realized this is impossible without some sort of a map. So I created FiveMap, a matrix the same size of FiveCheck, with the same formulas except the word "board" is removed. For example:
FiveMap(34,:) = [(7),(14),(21),(28),(35)];
FiveMap(35,:) = [(14),(21),(28),(35),(42)];
Since you are dealing with binary vectors of size 5, a very efficient solution might be the use of a look up table.
Consider board to be a binary matrix. you can filter it with 4 filters (of length 5), representing horizontal, vertical and two diagonals to identify possible positions you are looking for. Then, for specious locations, you can extract the 5 binary bits and use a look up table of size 32 to get the offset to the position where you should place your piece.
a small example:
% construct LUT
LUT = zeros(32,2); % dx and dy offsets for each entry
LUT(12,:) = [ 1 0 ]; % covering the case [0 1 1 0 0] - put piece 1 to the right of center
% continue constructing LUT here...
horFilt = ones(1, 5);
resp = imfilter( board, horFilt ); % need to think about board''s boundaries here...
[yy xx] = find( resp == 2 ); all locations where filter caught 2 out of 5
pat = board( yy, xx + [ -2 1 0 1 2] ); % I assume here only one location was found
pat = bin2dec( '0'+char( pat ) ); % convert binary pattern to decimal entry
board( yy + LUT(pat,2) , xx + LUT(pat, 1) ) = ... ; % your move here...