New answers tagged

0

The reason they are different is because (1) your eye have different sensitivity to different colors, and (b) display devices have very different color output. Therefore the luminescence need to be adjusted accordingly. Just adding the references and corrected number (0.7152): # 0.2989 *R + 0.5870 *G + 0.1140 *B # [1] NTSC # 0.2126 *R + 0.7152 *G + 0....


1

Feature Extraction There are many modern known features for images. Among them: BRISK Feature. FAST Feature. Harris Feature. KAZE Feature. MSER Feature. ORB Feature. SIFT Feature. SURF Feature. LBP Feature. HOG Features. Those are classic and popular features. Since the blossom of Deep Learning people are less and less invest in researching newer features. ...


0

I suggest you take a look at this paper: End-to-End Learning for Joint Image Demosaicing, Denoising and Super-Resolution. in Section 3. Proposed method, they have discussed the execution order of image demosaicing, denoising, and super-resolution. From the paper, For the mixture problem of image demosaicing, denoising and super-resolution, a clean high-...


1

One way to interpret the DFT, that I personally find most useful, is by comparing it to the DTFT (Discrete Tome Fourier Transform, which has an infinite input domain) of a repeated function. That is, imagine tiling your image out to infinity, and applying the DTFT, to obtain a periodic, continuous Fourier domain that you then sample. Because of the ...


2

The Linear Discriminant Analysis (LDA) (Also the Fisher's Linear Discriminant, which the LDA is a generalization of) is a method to find a projection plane to separate data by linear projection Matrix multiplication). Its main limitation is the use of linear projection. On the other hand, it can be used in a supervised manner. Namely it can use the labels to ...


0

This is the important part: np.log10(np.abs(fshift0)). It does a logarithmic scaling of the magnitude of the complex spectrum. It is common to display the Fourier transform this way because of the large dynamic range. The rest is scaling the input to the log function, so you get a nice view of the data, and scaling its output, probably to get it in the 0-1 ...


2

This is probably going to take trial and error on your part. The things I'd try: Different light angles: use LEDs at different incident angles to illuminate the cards. Different light colors: Use red, green, blue, and infrared LEDs to illuminate the cards. Polarizing filter: use a polaroid filter (sunglasses might do) at different angles (rotations) in ...


0

If the image is a GIF (or similar), you have large areas a single colour. To find where these areas end, an edge detector is the logical choice. In DSP terms, look at high pass filters to detect changes when you move between areas. Note: jpeg and other picture formats can have areas where colour slowly changes. In this case, the edge between colours is not ...


0

Spectrograms will work with any network that can operate on images. A spectrogram, however, is not an image, and many image techniques will be inapplicable: Data augmentation via rotation: a rotated spectrogram doesn't represent the same process at all, or even any process (there may not be a signal that maps to a given 2D array). Some networks are tailored ...


-1

Y = fft2(x); imagesc(10*log10(abs(fftshift(Y))));


1

I'm very new to Fourier transform I suggest to start with the math here. Make sure you understand the formula and WHY it is the way it is. Then work your way up by doing simple examples where you write the code AND also calculate the result by hand so you can compare the two and get a feel of how the code works Start with a single non-zero pixel in the ...


3

You should try something like: F = fft2(img); figure; imagesc(abs(F)); In image processing many times we're after the Log Spectrum: F = fft2(img); figure; imagesc(log10(1 + abs(F)));


6

Unsupervised clustering of image data is tricky thing and requires adjusting the method to the content of the images set. Assuming we're dealing with the MNIST data set we can do some nice things using known tools. First, let's assume we're after 2 features, namely we're after a dimensionality reduction from 784 features / dimensions to 2. The first approach ...


2

I think that the best option, probably, would be Random Forests or any other Boosting / Bagging method based on decision trees. I would probably start with SK Learn Random Forests. As more advanced trick I'd go after XGBoost.


Top 50 recent answers are included