Search Results

Search found 42 results on 2 pages for 'convolution'.

Page 1/2 | 1 2  | Next Page >

  • CUDA small kernel 2d convolution - how to do it

    - by paulAl
    I've been experimenting with CUDA kernels for days to perform a fast 2D convolution between a 500x500 image (but I could also vary the dimensions) and a very small 2D kernel (a laplacian 2d kernel, so it's a 3x3 kernel.. too small to take a huge advantage with all the cuda threads). I created a CPU classic implementation (two for loops, as easy as you would think) and then I started creating CUDA kernels. After a few disappointing attempts to perform a faster convolution I ended up with this code: http://www.evl.uic.edu/sjames/cs525/final.html (see the Shared Memory section), it basically lets a 16x16 threads block load all the convolution data he needs in the shared memory and then performs the convolution. Nothing, the CPU is still a lot faster. I didn't try the FFT approach because the CUDA SDK states that it is efficient with large kernel sizes. Whether or not you read everything I wrote, my question is: how can I perform a fast 2D convolution between a relatively large image and a very small kernel (3x3) with CUDA?

    Read the article

  • convolution in R

    - by user236215
    I tried to do convolution in R directly and using FFTs then taking inverse. But it seems from simple observation it is not correct. Look at this example: # DIRECTLY > x2$xt [1] 24.610 24.605 24.610 24.605 24.610 > h2$xt [1] 0.003891051 0.003875910 0.003860829 0.003845806 0.003830842 > convolve(h2$xt,x2$xt) [1] 0.4750436 0.4750438 0.4750435 0.4750437 0.4750435 # USING INVERSE FOURIER TRANSFORM > f=fft(fft(h2$xt)*fft(x2$xt), inv=TRUE) > Re(f)/length(f) [1] 0.4750438 0.4750435 0.4750437 0.4750435 0.4750436 > Lets take the index 0. At 0, the convolution should simply be the last value of x2$xt (24.610) multiplied by first value of h2$xt (0.003891051) which should give convolution at index 0 = 24.610*0.003891051 = 0.09575877 which is way off from 0.4750436. Am I doing something wrong? Why is the values so different from expected?

    Read the article

  • 3d convolution in c++

    - by alboot
    Hello, I'm looking for some source code implementing 3d convolution. Ideally, I need C++ code or CUDA code. I'd appreciate if anybody can point me to a nice and fast implementation :-) Cheers

    Read the article

  • 2d convolution using python and numpy

    - by mikip
    Hi I am trying to perform a 2d convolution in python using numpy I have a 2d array as follows with kernel H_r for the rows and H_c for the columns data = np.zeros((nr, nc), dtype=np.float32) #fill array with some data here then convolve for r in range(nr): data[r,:] = np.convolve(data[r,:], H_r, 'same') for c in range(nc): data[:,c] = np.convolve(data[:,c], H_c, 'same') It does not produce the output that I was expecting, does this code look OK Thanks

    Read the article

  • Understanding The Convolution Matrix

    - by Ryan Naddy
    I am learning about the Convolution Matrix, and I understand how they work, but I don't understand how to know before hand what the output of a Matrix will look like. For example lets say I want to add a blur to an image, I could guess 10,000+ different combinations of numbers before I get the correct one. I do know though that this formula will give me a blur effect, but I have no idea why. float[] sharpen = new float[] { 1/9f, 1/9f, 1/9f, 1/9f, 1/9f, 1/9f, 1/9f, 1/9f, 1/9f }; Can anyone either explain to me how this works or point me to some article, that explains this? I would like to know before hand what a possible output of the matrix will be without guessing. Basically I would like to know why do we put that number in the filed, and why not some other number?

    Read the article

  • iPhone Image Processing--matrix convolution

    - by James
    I am implementing a matrix convolution blur on the iPhone. The following code converts the UIImage supplied as an argument of the blur function into a CGImageRef, and then stores the RGBA values in a standard C char array. CGImageRef imageRef = imgRef.CGImage; int width = imgRef.size.width; int height = imgRef.size.height; CGColorSpaceRef colorSpace = CGColorSpaceCreateDeviceRGB(); unsigned char *pixels = malloc((height) * (width) * 4); NSUInteger bytesPerPixel = 4; NSUInteger bytesPerRow = bytesPerPixel * (width); NSUInteger bitsPerComponent = 8; CGContextRef context = CGBitmapContextCreate(pixels, width, height, bitsPerComponent, bytesPerRow, colorSpace, kCGImageAlphaPremultipliedLast | kCGBitmapByteOrder32Big); CGContextDrawImage(context, CGRectMake(0, 0, width, height), imageRef); CGContextRelease(context); Then the pixels values stored in the pixels array are convolved, and stored in another array. unsigned char *results = malloc((height) * (width) * 4); Finally, these augmented pixel values are changed back into a CGImageRef, converted to a UIImage, and the returned at the end of the function with the following code. context = CGBitmapContextCreate(results, width, height, bitsPerComponent, bytesPerRow, colorSpace, kCGImageAlphaPremultipliedLast | kCGBitmapByteOrder32Big); CGImageRef finalImage = CGBitmapContextCreateImage(context); UIImage *newImage = [UIImage imageWithCGImage:CGBitmapContextCreateImage(context)]; CGImageRelease(finalImage); NSLog(@"edges found"); free(results); free(pixels); CGColorSpaceRelease(colorSpace); return newImage; This works perfectly, once. Then, once the image is put through the filter again, very odd, unprecedented pixel values representing input pixel values that don't exist, are returned. Is there any reason why this should work the first time, but then not afterward? Beneath is the entirety of the function. -(UIImage*) blur:(UIImage*)imgRef { CGImageRef imageRef = imgRef.CGImage; int width = imgRef.size.width; int height = imgRef.size.height; CGColorSpaceRef colorSpace = CGColorSpaceCreateDeviceRGB(); unsigned char *pixels = malloc((height) * (width) * 4); NSUInteger bytesPerPixel = 4; NSUInteger bytesPerRow = bytesPerPixel * (width); NSUInteger bitsPerComponent = 8; CGContextRef context = CGBitmapContextCreate(pixels, width, height, bitsPerComponent, bytesPerRow, colorSpace, kCGImageAlphaPremultipliedLast | kCGBitmapByteOrder32Big); CGContextDrawImage(context, CGRectMake(0, 0, width, height), imageRef); CGContextRelease(context); height = imgRef.size.height; width = imgRef.size.width; float matrix[] = {0,0,0,0,1,0,0,0,0}; float divisor = 1; float shift = 0; unsigned char *results = malloc((height) * (width) * 4); for(int y = 1; y < height; y++){ for(int x = 1; x < width; x++){ float red = 0; float green = 0; float blue = 0; int multiplier=1; if(y>0 && x>0){ int index = (y-1)*width + x; red = matrix[0]*multiplier*(float)pixels[4*(index-1)] + matrix[1]*multiplier*(float)pixels[4*(index)] + matrix[2]*multiplier*(float)pixels[4*(index+1)]; green = matrix[0]*multiplier*(float)pixels[4*(index-1)+1] + matrix[1]*multiplier*(float)pixels[4*(index)+1] + matrix[2]*multiplier*(float)pixels[4*(index+1)+1]; blue = matrix[0]*multiplier*(float)pixels[4*(index-1)+2] + matrix[1]*multiplier*(float)pixels[4*(index)+2] + matrix[2]*multiplier*(float)pixels[4*(index+1)+2]; index = (y)*width + x; red = red+ matrix[3]*multiplier*(float)pixels[4*(index-1)] + matrix[4]*multiplier*(float)pixels[4*(index)] + matrix[5]*multiplier*(float)pixels[4*(index+1)]; green = green + matrix[3]*multiplier*(float)pixels[4*(index-1)+1] + matrix[4]*multiplier*(float)pixels[4*(index)+1] + matrix[5]*multiplier*(float)pixels[4*(index+1)+1]; blue = blue + matrix[3]*multiplier*(float)pixels[4*(index-1)+2] + matrix[4]*multiplier*(float)pixels[4*(index)+2] + matrix[5]*multiplier*(float)pixels[4*(index+1)+2]; index = (y+1)*width + x; red = red+ matrix[6]*multiplier*(float)pixels[4*(index-1)] + matrix[7]*multiplier*(float)pixels[4*(index)] + matrix[8]*multiplier*(float)pixels[4*(index+1)]; green = green + matrix[6]*multiplier*(float)pixels[4*(index-1)+1] + matrix[7]*multiplier*(float)pixels[4*(index)+1] + matrix[8]*multiplier*(float)pixels[4*(index+1)+1]; blue = blue + matrix[6]*multiplier*(float)pixels[4*(index-1)+2] + matrix[7]*multiplier*(float)pixels[4*(index)+2] + matrix[8]*multiplier*(float)pixels[4*(index+1)+2]; red = red/divisor+shift; green = green/divisor+shift; blue = blue/divisor+shift; if(red<0){ red=0; } if(green<0){ green=0; } if(blue<0){ blue=0; } if(red>255){ red=255; } if(green>255){ green=255; } if(blue>255){ blue=255; } int realPos = 4*(y*imgRef.size.width + x); results[realPos] = red; results[realPos + 1] = green; results[realPos + 2] = blue; results[realPos + 3] = 1; }else { int realPos = 4*((y)*(imgRef.size.width) + (x)); results[realPos] = 0; results[realPos + 1] = 0; results[realPos + 2] = 0; results[realPos + 3] = 1; } } } context = CGBitmapContextCreate(results, width, height, bitsPerComponent, bytesPerRow, colorSpace, kCGImageAlphaPremultipliedLast | kCGBitmapByteOrder32Big); CGImageRef finalImage = CGBitmapContextCreateImage(context); UIImage *newImage = [UIImage imageWithCGImage:CGBitmapContextCreateImage(context)]; CGImageRelease(finalImage); free(results); free(pixels); CGColorSpaceRelease(colorSpace); return newImage;} THANKS!!!

    Read the article

  • Convolving two signals

    - by John Elway
    Calculate the convolution of the following signals (your answer will be in the form of an equation): h[n] = delta[n-1] + delta[n+1], x[n] = delta[n-a] + delta[n+b] I'm lost as to what I do with h and x. Do I simply multiply them? h[n]*x[n]? I programmed convolution with several types of blurs and edge detectors, but I don't see how to translate that knowledge to this problem. please help!

    Read the article

  • OpenMP + SSE gives no speedup

    - by Sayan Ghosh
    Hi, My Professor found out this interesting experiment of 3D Linearly separable Kernel Convolution using SSE and OpenMP, and gave the task to me to benchmark the statistics on our system. The author claims a crazy 18 fold speedup from the serial approach! Might not be always, but we were expecting at least a 2-4 times speedup running this on a Dual Core Intel. http://software.intel.com/en-us/articles/16bit-3d-convolution-sse4openmp-implementation-on-penryn-cpu/#comment-41994 Alas, we could find exactly no speedup. The serial code performs always better, with or without OpenMP. I am using Linux, and observed a certain trend...when no other processes are running on the system, after a while the loadavg starts increasing, and the the %CPU utilization falls down. Another probable false positive which I ran into accidentally...I started the program, then immediately paused it. Then I ran it on background with bg, and saw a speedup of more than 2. This happens all the time! Any advice would be great. Thanks, Sayan

    Read the article

  • My OpenCL kernel is slower on faster hardware.. But why?

    - by matdumsa
    Hi folks, As I was finishing coding my project for a multicore programming class I came up upon something really weird I wanted to discuss with you. We were asked to create any program that would show significant improvement in being programmed for a multi-core platform. I’ve decided to try and code something on the GPU to try out OpenCL. I’ve chosen the matrix convolution problem since I’m quite familiar with it (I’ve parallelized it before with open_mpi with great speedup for large images). So here it is, I select a large GIF file (2.5 MB) [2816X2112] and I run the sequential version (original code) and I get an average of 15.3 seconds. I then run the new OpenCL version I just wrote on my MBP integrated GeForce 9400M and I get timings of 1.26s in average.. So far so good, it’s a speedup of 12X!! But now I go in my energy saver panel to turn on the “Graphic Performance Mode” That mode turns off the GeForce 9400M and turns on the Geforce 9600M GT my system has. Apple says this card is twice as fast as the integrated one. Guess what, my timing using the kick-ass graphic card are 3.2 seconds in average… My 9600M GT seems to be more than two times slower than the 9400M.. For those of you that are OpenCL inclined, I copy all data to remote buffers before starting, so the actual computation doesn’t require roundtrip to main ram. Also, I let OpenCL determine the optimal local-worksize as I’ve read they’ve done a pretty good implementation at figuring that parameter out.. Anyone has a clue? edit: full source code with makefiles here http://www.mathieusavard.info/convolution.zip cd gimage make cd ../clconvolute make put a large input.gif in clconvolute and run it to see results

    Read the article

  • Image processing custom filter 7 by 7

    - by ladiesMan217
    Lets say I have a 7 by 7 neighborhood around a pixel that looks like this 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 and I wanna filter the above by replacing the pixel p by the average of those pixels whose value lie in the range -10<=p_value <=10. I am new to image processing and I think in this case p_value is 25 and around 25 that are many pixel values in that range but don't exactly know to construct a convolution filter out of it.

    Read the article

  • Small-o(n^2) implementation of Polynomial Multiplication

    - by AlanTuring
    I'm having a little trouble with this problem that is listed at the back of my book, i'm currently in the middle of test prep but i can't seem to locate anything regarding this in the book. Anyone got an idea? A real polynomial of degree n is a function of the form f(x)=a(n)x^n+?+a1x+a0, where an,…,a1,a0 are real numbers. In computational situations, such a polynomial is represented by a sequence of its coefficients (a0,a1,…,an). Assuming that any two real numbers can be added/multiplied in O(1) time, design an o(n^2)-time algorithm to compute, given two real polynomials f(x) and g(x) both of degree n, the product h(x)=f(x)g(x). Your algorithm should **not** be based on the Fast Fourier Transform (FFT) technique. Please note it needs to be small-o(n^2), which means it complexity must be sub-quadratic. The obvious solution that i have been finding is indeed the FFT, but of course i can't use that. There is another method that i have found called convolution, where if you take polynomial A to be a signal and polynomial B to be a filter. A passed through B yields a shifted signal that has been "smoothed" by A and the resultant is A*B. This is supposed to work in O(n log n) time. Of course i am completely unsure of implementation. If anyone has any ideas of how to achieve a small-o(n^2) implementation please do share, thanks.

    Read the article

  • Multi-dimensional array edge/border conditions

    - by kirbuchi
    Hi, I'm iterating over a 3 dimensional array (which is an image with 3 values for each pixel) to apply a 3x3 filter to each pixel as follows: //For each value on the image for (i=0;i<3*width*height;i++){ //For each filter value for (j=0;j<9;j++){ if (notOutsideEdgesCondition){ *(**(outArray)+i)+= *(**(pixelArray)+i-1+(j%3)) * (*(filter+j)); } } } I'm using pointer arithmetic because if I used array notation I'd have 4 loops and I'm trying to have the least possible number of loops. My problem is my notOutsideEdgesCondition is getting quite out of hands because I have to consider 8 border cases. I have the following handled conditions Left Column: ((i%width)==0) && (j%3==0) Right Column: ((i-1)%width ==0) && (i>1) && (j%3==2) Upper Row: (i<width) && (j<2) Lower Row: (i>(width*height-width)) && (j>5) and still have to consider the 4 corner cases which will have longer expressions. At this point I've stopped and asked myself if this is the best way to go because If I have a 5 line long conditional evaluation it'll not only be truly painful to debug but will slow the inner loop. That's why I come to you to ask if there's a known algorithm to handle this cases or if there's a better approach for my problem. Thanks a lot.

    Read the article

  • Mulit-dimensional array edge/border conditions

    - by kirbuchi
    Hi, I'm iterating over a 3 dimensional array (which is an image with 3 values for each pixel) to apply a 3x3 filter to each pixel as follows: //For each value on the image for (i=0;i<3*width*height;i++){ //For each filter value for (j=0;j<9;j++){ if (notOutsideEdgesCondition){ *(**(outArray)+i)+= *(**(pixelArray)+i-1+(j%3)) * (*(filter+j)); } } } I'm using pointer arithmetic because if I used array notation I'd have 4 loops and I'm trying to have the least possible number of loops. My problem is my notOutsideEdgesCondition is getting quite out of hands because I have to consider 8 border cases. I have the following handled conditions Left Column: ((i%width)==0) && (j%3==0) Right Column: ((i-1)%width ==0) && (i>1) && (j%3==2) Upper Row: (i<width) && (j<2) Lower Row: (i>(width*height-width)) && (j>5) and still have to consider the 4 corner cases which will have longer expressions. At this point I've stopped and asked myself if this is the best way to go because If I have a 5 line long conditional evaluation it'll not only be truly painful to debug but will slow the inner loop. That's why I come to you to ask if there's a known algorithm to handle this cases or if there's a better approach for my problem. Thanks a lot.

    Read the article

  • How to remove the boundary effects arising due to zero padding in scipy/numpy fft?

    - by Omkar
    I have made a python code to smoothen a given signal using the Weierstrass transform, which is basically the convolution of a normalised gaussian with a signal. The code is as follows: #Importing relevant libraries from __future__ import division from scipy.signal import fftconvolve import numpy as np def smooth_func(sig, x, t= 0.002): N = len(x) x1 = x[-1] x0 = x[0] # defining a new array y which is symmetric around zero, to make the gaussian symmetric. y = np.linspace(-(x1-x0)/2, (x1-x0)/2, N) #gaussian centered around zero. gaus = np.exp(-y**(2)/t) #using fftconvolve to speed up the convolution; gaus.sum() is the normalization constant. return fftconvolve(sig, gaus/gaus.sum(), mode='same') If I run this code for say a step function, it smoothens the corner, but at the boundary it interprets another corner and smoothens that too, as a result giving unnecessary behaviour at the boundary. I explain this with a figure shown in the link below. Boundary effects This problem does not arise if we directly integrate to find convolution. Hence the problem is not in Weierstrass transform, and hence the problem is in the fftconvolve function of scipy. To understand why this problem arises we first need to understand the working of fftconvolve in scipy. The fftconvolve function basically uses the convolution theorem to speed up the computation. In short it says: convolution(int1,int2)=ifft(fft(int1)*fft(int2)) If we directly apply this theorem we dont get the desired result. To get the desired result we need to take the fft on a array double the size of max(int1,int2). But this leads to the undesired boundary effects. This is because in the fft code, if size(int) is greater than the size(over which to take fft) it zero pads the input and then takes the fft. This zero padding is exactly what is responsible for the undesired boundary effects. Can you suggest a way to remove this boundary effects? I have tried to remove it by a simple trick. After smoothening the function I am compairing the value of the smoothened signal with the original signal near the boundaries and if they dont match I replace the value of the smoothened func with the input signal at that point. It is as follows: i = 0 eps=1e-3 while abs(smooth[i]-sig[i])> eps: #compairing the signals on the left boundary smooth[i] = sig[i] i = i + 1 j = -1 while abs(smooth[j]-sig[j])> eps: # compairing on the right boundary. smooth[j] = sig[j] j = j - 1 There is a problem with this method, because of using an epsilon there are small jumps in the smoothened function, as shown below: jumps in the smooth func Can there be any changes made in the above method to solve this boundary problem?

    Read the article

  • GLSL - one-pass gaussian blur

    - by martin pilch
    It is possible to implement fragment shader to do one-pass gaussian blur? I have found lot of implementation of two-pass blur (gaussian and box blur): http://callumhay.blogspot.com/2010/09/gaussian-blur-shader-glsl.html http://www.gamerendering.com/2008/10/11/gaussian-blur-filter-shader/ http://www.geeks3d.com/20100909/shader-library-gaussian-blur-post-processing-filter-in-glsl/ and so on. I have been thinking of implementing gaussian blur as convolution (in fact, it is the convolution, the examples above are just aproximations): http://en.wikipedia.org/wiki/Gaussian_blur

    Read the article

  • Search Display Controller Crashes When Returning Results

    - by Convolution
    I have a tableview with a search display controller. It has been working fine in the past, but recently has started crashing for certain search results. Here my code searches a Golfer based on their Name, Age and Handicap. The data is correctly loaded into the table, I can access and drill down to receive further information. However when I type in a search query for either Name or Age, the app crashes, while the Golfers Handicap is returned fine. Note: dataSouceArray is the data source for the tableview, dataSourceArrayCopy is the mutable copy of the data used to add and remove objects in the search filter. - (void)filterContentForSearchText:(NSString*)searchText scope:(NSString*)scope { /* Update the filtered array based on the search text and scope. */ [self.dataSourceArrayCopy removeAllObjects]; // First clear the filtered array. /* Search the main list for products whose type matches the scope (if selected) and whose name matches searchText; add items that match to the filtered array. */ for (Golfer *golfer in dataSourceArray) { if ([scope isEqualToString:@"Name"] || [golfer.golferName isEqualToString:scope]) { NSComparisonResult result = [golfer.golferName compare:searchText options:(NSCaseInsensitiveSearch|NSDiacriticInsensitiveSearch) range:NSMakeRange(0, [searchText length])]; if (result == NSOrderedSame) { [self.customerListCopy addObject:golfer]; } } if ([scope isEqualToString:@"Age"] || [golfer.golferAge isEqualToString:scope]) { NSComparisonResult result = [golfer.golferAge compare:searchText options:(NSCaseInsensitiveSearch|NSDiacriticInsensitiveSearch) range:NSMakeRange(0, [searchText length])]; if (result == NSOrderedSame) { [self.dataSourceArrayCopy addObject:golfer]; } } if ([scope isEqualToString:@"Handicap"] || [golfer.golferHandicap isEqualToString:scope]) { NSComparisonResult result = [golfer.golferHandicap compare:searchText options:(NSCaseInsensitiveSearch|NSDiacriticInsensitiveSearch) range:NSMakeRange(0, [searchText length])]; if (result == NSOrderedSame) { [self.dataSourceArrayCopy addObject:golfer]; } } } } - (BOOL)searchDisplayController:(UISearchDisplayController *)controller shouldReloadTableForSearchString:(NSString *)searchString { [self filterContentForSearchText:searchString scope: [[self.searchDisplayController.searchBar scopeButtonTitles] objectAtIndex:[self.searchDisplayController.searchBar selectedScopeButtonIndex]]]; // Return YES to cause the search result table view to be reloaded. return YES; } - (BOOL)searchDisplayController:(UISearchDisplayController *)controller shouldReloadTableForSearchScope:(NSInteger)searchOption { [self filterContentForSearchText:[self.searchDisplayController.searchBar text] scope: [[self.searchDisplayController.searchBar scopeButtonTitles] objectAtIndex:searchOption]]; // Return YES to cause the search result table view to be reloaded. return YES; } Any help would be appreciated, thank you for taking the time to read this.

    Read the article

  • Filtering spectrum using FIR filters

    - by Alex Hoppus
    If i have signal values x[T] and filter coefficients b[i], i can perform filtering using convolution. Suppose i have spectrum of x (after FFT) and i need to perform filtering using filters coefficients, how can i perform this? I heard that in frequency domain it will be multiplying, rather than convolution (time domain). But i can't find an equation to use it. I have 614000 values in y = fft(x[T]) vector and 119 filter coefficients (generated using fdatool), i can't multiply them directly ... Thanks.

    Read the article

  • Setting pixel values in Nvidia NPP ImageCPU objects?

    - by solvingPuzzles
    In the Nvidia Performance Primitives (NPP) image processing examples in the CUDA SDK distribution, images are typically stored on the CPU as ImageCPU objects, and images are stored on the GPU as ImageNPP objects. boxFilterNPP.cpp is an example from the CUDA SDK that uses these ImageCPU and ImageNPP objects. When using a filter (convolution) function like nppiFilter, it makes sense to define a filter as an ImageCPU object. However, I see no clear way setting the values of an ImageCPU object. npp::ImageCPU_32f_C1 hostKernel(3,3); //allocate space for 3x3 convolution kernel //want to set hostKernel to [-1 0 1; -1 0 1; -1 0 1] hostKernel[0][0] = -1; //this doesn't compile hostKernel(0,0) = -1; //this doesn't compile hostKernel.at(0,0) = -1; //this doesn't compile How can I manually put values into an ImageCPU object? Notes: I didn't actually use nppiFilter in the code snippet; I'm just mentioning nppiFilter as a motivating example for writing values into an ImageCPU object. The boxFilterNPP.cpp example doesn't involve writing directly to an ImageCPU object, because nppiFilterBox is a special case of nppiFilter that uses a built-in gaussian smoothing filter (probably something like [1 1 1; 1 1 1; 1 1 1]).

    Read the article

  • Face Recognition for classifying digital photos?

    - by Jeremy E
    I like to mess around with AI and wanted to try my hand at face recognition the first step is to find the faces in the photographs. How is this usually done? Do you use convolution of a sample image/images or statistics based methods? How do you find the bounding box for the face? My goal is to classify the pictures of my kids from all the digital photos. Thanks in advance.

    Read the article

  • String munging in Objective-C with NSAttributedString.

    - by dreeves
    I have an NSAttributedString s and an integer i and I'd like a function that takes s and i and returns a new NSAttributedString that has a (stringified) i prepended to s. It looks like some combination of -stringWithFormat:, -initWithString:, and -insertAttributedString: would do it but I'm having trouble piecing it together without a lot of convolution and temporary variables. More generally, pointers to guides on making sense of NSAttributedString and NSMutableAttributedString would be awesome.

    Read the article

  • mean image filter

    - by turmoil
    Starting to learn image filtering and stumped on a question found on website: Applying a 3×3 mean filter twice does not produce quite the same result as applying a 5×5 mean filter once. However, a 5×5 convolution kernel can be constructed which is equivalent. What does this kernel look like? Would appreciate help so that I can understand the subject better. Thanks.

    Read the article

  • High speed matrix manipulation in c#?

    - by Donnie
    I'm working on some image manipulation code in c# and need to do some matrix operations (specifically 2D convolution). I have the code written in matlab which uses the conv2 function ... is there a library for C# / .NET that does good high-speed matrix manipulations? I'd be fine if it requires some specific GPU and does the matrix math on-GPU if that's what it takes.

    Read the article

  • Subband decomposition using Daubechies filter

    - by misha
    I have the following two 8-tap filters: h0 ['-0.010597', '0.032883', '0.030841', '-0.187035', '-0.027984', '0.630881', '0.714847', '0.230378'] h1 ['-0.230378', '0.714847', '-0.630881', '-0.027984', '0.187035', '0.030841', '-0.032883', '-0.010597'] Here they are on a graph: I'm using it to obtain the approximation (lower subband of an image). This is a(m,n) in the following diagram: I got the coefficients and diagram from the book Digital Image Processing, 3rd Edition, so I trust that they are correct. The star symbol denotes one dimensional convolution (either over rows or over columns). The down arrow denotes downsampling in one dimension (either over rows, or columns). My problem is that the filter coefficients for h0 and h1 sum to greater than 1 (approximately 1.4 or sqrt(2) to be exact). Naturally, if I convolve any image with the filter, the image will get brighter. Indeed, here's what I get (expected result on right): Can somebody suggest what the problem is here? Why should it work if the convolution filter coefficients sum to greater than 1? I have the source code, but it's quite long so I'm hoping to avoid posting it here. If it's absolutely necessary, I'll put it up later. EDIT What I'm doing is: Decompose into subbands Filter one of the subbands Recompose subbands into original image Note that the point isn't just to have a displayable subband-decomposed image -- I have to be able to perfectly reconstruct the original image from the subbands as well. So if I scale the filtered image in order to compensate for my decomposition filter making the image brighter, this is what I will have to do: Decompose into subbands Apply intensity scaling Filter one of the subbands Apply inverse intensity scaling Recompose subbands into original image Step 2 performs the scaling. This is what @Benjamin is suggesting. The problem is that then step 4 becomes necessary, or the original image will not be properly reconstructed. This longer method will work. However, the textbook explicitly says that no scaling is performed on the approximation subband. Of course, it's possible that the textbook is wrong. However, what's more possible is I'm misunderstanding something about the way this all works -- this is why I'm asking this question.

    Read the article

  • Using textureGrad for anisotropic integration approximation

    - by Amxx
    I'm trying to develop a real time rendering method using real time acquired envmap (cubemap) for lightning. This implies that my envmap can change as often as every frame and I therefore cannot use any method base on precomputation of the envmap (such as convolution with BRDF...) So far my method worked well with Phong BRDF. For specular contribution I direclty read the value in my sampleCube and I use mipmap levels + linear filter for simulating the roughtness of the material considered: int size = textureSize(envmap, 0).x; float specular_level = log2(size * sqrt(3.0)) - 0.5 * log2(ns + 1); vec3 env_specular = ks * specular_color * textureLod(envmap, l_g, specular_level); From this method I would like to upgrade to a microfacet based BRDF. I already have algorithm for evaluating the shape (including anisotropic direction) of the reflection but I cannot manage to read the values I want in my sampleCube. I believe I have to use textureGrad(envmap, l_g, X, Y); with l_g being the reflection direction in global space but I cannot manage to find which values to give to X and Y in order to correctly specify the area I want to consider. What value should I give to X and Y in orther for textureGrad(envmap, l_g, X, Y); to give the same result as textureLod(envmap, l_g, specular_level);

    Read the article

1 2  | Next Page >