Feature Detection And Matching

How Outliers are Removed? 

The matched features between the images have a euclidean distance. Features with euclidean distance beyond the average euclidean or a specified threshold are removed as outliers.

SIFT(Lowe, 2004):

Detects special features regardless of the scale and orientation of the image, and allows you to reliably detect the same special features even in slightly distorted images, adding noise or changing the lighting and / or viewing point. SIFT detects potentially specific features, and measures the stability of these properties and determines their magnitude by eliminating unstable properties. Then, according to the local gradient direction, for each specific feature, one or more orientations are calculated and assigned. With this information, image data properties can be normalized to scale, position and orientation - so the properties become scalable with respect to these transformations. The method also includes a descriptor, which detects the special properties described by the 128-dimensional vectors. The gradient values and orientations are initially calculated for the position of the surrounding object, using the Gaussian filter for the entire image. Then the descriptor's coordinates and gradient orientations are rotated before they are detected in the direction of the special properties. So the descriptor maintains a normalized orientation of the special characteristic.

SURF (Bay et al., 2006):

Uses second-order Gaussian derivative approximations with a box filter, thus losing some accuracy, but significantly shortening the calculation time. Detecting properties at different image scales, in contrast to SIFT, does not need to use a Gaussian filter, but it is enough to change the size of the box used, again avoiding expensive time calculations. SURF descriptors only use the 64-dimensional vector, which is easier to generate and compare, but saves less information that may be useful in itself.

FAST is a method for corner detection (Rosten and Drummond, 2006). The main feature of this detector is the speed. The FAST method is available in real-time (using only 7% of the time for single- frame processing) to detect corners in a PAL format video. The algorithm is characterized by speed, but is not resistant to large noise quantities in pictures, and results depend on the choice of threshold value.

MSER (Matas et al., 2004) detects specific regions of an image, which form a set of interconnected image points that make up the contour after the thresholding of the image. The intensity of all points within these regions is either lighter or darker than the points on the contour. Such regions are invariant to scaling, lighting, orientation and viewing point transforms.

Comments

Post a Comment

Popular posts from this blog

Electrophoretic Display

Image
Electrophoresis is the motion of dispersed particles relative to a fluid under the influence of a spatially uniform electric field. E Ink Corporation further describes their electronic ink technology as a type of display made of millions of tiny and transparent microcapsules—measuring about the diameter of a human hair. Each microcapsule represents a pixel and each contains a positively charged white pigment particle and a negatively charged black pigment particle. Upon the application of positive or negative electric, corresponding particles move to the top of the microcapsule where they become visible to the user. This makes the surface appear white or black at that spot. Advantages: 1) Lower Power Consumption: Note that this display technology is a bistable display. This means that the image on an electrophoretic display screen is retained even when all power sources are removed. The display is only consuming power whenever the displayed image changes. 2) Refl...
Read more

Color Electrophoretic Display

Image
Until now, the only way to get "full" color from an EPD -- at least the only way that E Ink has shown us -- is placing a matrix color filter in front of the monochrome display. E Ink showed the way forward a few years ago with a black, white, and red display, which managed to control particles of three different colors using differences in mobility and a cleverly designed controlling waveform. At Display Week 2016, E Ink introduced an impressive expansion of this approach, in which particles of four different colors are included within each microcapsule, given different mobilities through different sizing, and driven with a pulsed controlling wave movement that permits the creation of thousands of colors, as explained by E Ink's Giovanni Mancini.     The resulting display showed impressively bright and saturated colors and drew crowds.  When a new image was written, the display would flash several times.  It took about 10 seconds for a new image to build...
Read more

Histogram Equalization Algorithm Implementation Using MATLAB

Image
  Histogram equalization is a method in image processing of contrast adjustment using the image's histogram. The histogram of a digital image is a distribution of its discrete intensity levels in the range [0,L-1]. The distribution is a discrete function h associating to each intensity level: r k the number of pixel with this intensity: n k . Normalization of a Histogram Normalize an histogram is a technique consisting into transforming the discrete distribution of intensities into a discrete distribution of probabilities. To do so, we need to divide each value of the histogram by the number of pixel. Because a digital image is a discrete set of values that could be seen as a matrix and it's equivalent to divide each n k by the dimension of the array which is the product of the width by the length of the image.   Equalization of a Histogram Histogram equalization is a method to process images in order to adjust the contrast of an image by modifyin...
Read more