Pattern recognition (psychology)

Feature (linguistics)

Noise (video)

Gene

Image (mathematics)

Feature detection (computer vision)

Philosophy

Histogram of oriented gradients

Artificial intelligence

Speckle noise

Image gradient

Chemistry

Transformation (genetics)

Computer science

Linguistics

Feature extraction

Histogram

Scale-invariant feature transform

Computer vision

Biochemistry

Image texture

Image processing

We present the performance of three popular image feature extraction methods such as Scale Invariant Feature Transformation (SIFT), Speeded-Up Robust Features (SURF) and Histogram of Oriented Gradient (HOG). Specifically, we compare the performance of feature detection methods for images corrupted with different types of noise. The efficiency of three methods are measured by considering number of correct matches between original and noisy image found by the algorithm. In this study, we use images corrupted by three types of noise such as gaussian, salt & pepper and speckle. It is observed from the experimental results that for most of the noisy images, SIFT presents its stability but it is slow. SURF is the fastest one and its performance is close to SIFT. However, HOG show its advantages in detecting edge and texture information of image.

Analysis of various image feature extraction methods against noisy image: SIFT, SURF and HOG