Matlab recursively add points to a vector

Hi, I have a 3d matrix and i wish to create a function that looks at its enviroment and if a point in that enviroment satisfies a condition i want to add it to my vector. then use the function on that point to see if it also satisfies and so on. my idea was to preform a recursive function that calls itself on the next point that satisfies the condition. the only problem is that for every point i add to the vector Matlab would copy the entire vector array, this will slow me down incredibly as that vector is incredibly large, from what i understand passing a pointer to the array is impossible in matlab so im confused as to how am i to do it?

my question is mostly how do i pass an array of vectors as pointer so that matlab keeps feeling it instead of re-creating it over and over again.

[Jan]

I do not understand the detail concerning "looks at its environment". What does this mean? What exactly is a "vector array"? A matrix? Matlab uses a copy-on-write strategy, such that you do not create a deep data copy, when a variable is provided as input to a function. Only the iterative growing of arrays is a problem. Terms like "incredibly large" are not useful in a scientific discussion. Some people are impressed by 10'000 elements already, because they cannot add them manually anymore. Other users run Matlab on huge TB cluster machines.

Answer 1

[Matt J]

I would recommend a different approach entirely. I would first evaluate every voxel (i,j,k) in the array to see if it satisfies the condition. This will give you a 3D binary map BW. Then, I would use bwconncomp or regionprops in the Image Processing ToolBox to find the groups of 'on' voxels in BW that are connected through a chain of neighbors.

There are many possible variations on this idea that could be applicable if you give a more detailed description of the task.

[Daniel ziv]

Well i have a 3d image of cells with long strands and I've found seeds that represent those cells centroids, now i'm trying to create a function that spreads from the seed as long as the environments fluorescence is larger then a threshold. if i use a recursion that spreads in all direction the computation should be lower then O(n) where n are the amount of pixels which satisfy the condition. sadly simply finding the objects will not be satisfactory as the seeds are found using other channels and i only want the cells that sorround the seeds.

[Daniel ziv]

actually it seems very possible if i locally threshold my cells to turn them into binary and then use connected_components and then check if my seed hits the connected component. I find it hard to find all the methods nessasary if its simple enough can u plz help me with those few lines of code?

[Image Analyst]

You forgot to attach the image. Even a 2-D slice would help me to understand. I'm inclined to agree with Matt J so far, but then, we haven't seen your image yet.

Answer 2

[Jan]

See Lorens Blog: Inplace operations . Maybe your assumption is wrong, that Matlab copies the "vector array" (what ever this means) in each call of a function. If you explain, how many elements "incredibly large" is, one could estimate, if a cell array is a useful solution for you. So please add more details and explain, why you think that the memory management is the bottleneck of your code.