Up: Component summary
Up: Component summary
Component
TextureFeaturesThe component uses VLFeat library to extract SIFT/MSER keypoints and SIFT descriptors from grayscale images. The descriptors are used in texture analysis. In order to use this component, you have to get VLFEat (>=0.99) and extract it under the bundle folder so, that the vl_setup.m file will be in location: [BUNDLE_HOME]/lib/vlfeat/toolbox/vl_setup.m.
The "sift" method searches for keypoints with given parameters and produces the descriptors for those. The "dsift" or dense sift method extracts descriptors over a grid of keypoints. The "mser" method classifies the pixels and produces a segmented image and keypoint locations.
| Version | 1.0 |
|---|---|
| Bundle | anima |
| Categories | Image Analysis |
| Authors | Ville Rantanen (ville.rantanen@helsinki.fi) |
| Issue tracker | View/Report issues |
| Requires | Matlab ; VLFeat ; download (bash) |
| Source files | component.xml imagevlfeat.m pixelframe.m execute.m |
| Usage | Example with default values |
Inputs| Name | Type | Mandatory | Description |
|---|---|---|---|
| in | ImageList | Mandatory | Source grayscale images |
| names | ImageList | Optional | Source images for naming of the data |
Outputs| Name | Type | Description |
|---|---|---|
| images | ImageList | Result images, or depending on the method, visualizations. |
| out | CSVList | Result features and keypoints, separate CSV file for each input image file. |
Parameters| Name | Type | Default | Description |
|---|---|---|---|
| dsiftStep | float | 1 | Extracts a SIFT descriptor each STEP pixels. Method: dsift |
| method | string | "sift" | Method to use sift,dsift or mser |
| mserBrightOnDark | boolean | true | Detect bright-on-dark MSERs. This corresponds to MSERs of the inverted image. Method: mser |
| mserDarkOnBright | boolean | true | Detect dark-on-bright MSERs. This corresponds to MSERs of the original image. Method: mser |
| mserDelta | float | 5 | DELTA parameter of the MSER algorithm. Roughly speaking, the stability of a region is the relative variation of the region area when the intensity is changed of +/- Delta/2 (in UINT8). Method: mser |
| mserDraw | string | "regions" | Choices: regions, ellipses. Output image has the regions with overlaps, or the fitted ellipses on the regions. Method: mser |
| mserMaxArea | float | 0.75 | Set the maximum area of the regions relative to the image domain area. Method: mser |
| mserMaxVariation | float | 0.25 | Set the maximum variation (absolute stability score) of the regions. Method: mser |
| mserMinArea | float | 0 | Set the minimum area of the regions relative to the image domain area. Method: mser |
| mserMinDiversity | float | 0.2 | Set the minimum diversity of the region. When the relative area variation of two nested regions is below this threshold, then only the most stable one is selected. Method: mser |
| siftEdgeThresh | float | 10 | Non-edge selection threshold. Method: sift |
| siftFirstOctave | float | 0 | Index of the first octave of the DoG scale space. Method: sift |
| siftLevels | float | 3 | Number of levels per octave of the DoG scale space. Method: sift |
| siftMagnif | float | 3 | Descriptor magnification factor. The scale of the keypoint is multiplied by this factor to obtain the width (in pixels) of the spatial bins. Method: sift,dsift |
| siftOctaves | float | 0 | Number of octaves of the DoG scale space, defaults to "max possible". Method: sift |
| siftPeakThresh | float | 0 | Peak selection threshold. Method: sift |
| siftSize | float | 2 | The variance of the Gaussian window that determines the descriptor support. It is expressend in units of spatial bins. Method: sift,dsift |
Test cases| Test case | Parameters▼ | IN in |
IN names |
OUT images |
OUT out |
|
|---|---|---|---|---|---|---|
| case1_sift | properties | in | (missing) | images | out | |
|
metadata.timeout=60 |
||||||
| case2_mser | properties | in | (missing) | images | out | |
|
method=mser, |
||||||