G06V 10/40
Definition
Diese Klassifikationsstelle umfasst:(Für diese Definition ist die deutsche Übersetzung noch nicht abgeschlossen)
Methods and arrangements for extracting visual features which are subsequently input to an object recognition algorithm
Notes – technical background
These notes provide more information about the technical subject matter that is classified in this place:
Formerly, the choice of suitable feature extraction algorithms was a crucial design choice in the art of pattern recognition algorithms. It had a strong influence of the overall performance. With the advent of deep learning, particularly in convolutional neural networks, the need for the hand-picked design of dedicated feature extraction algorithms has decreased to some extent, because the inner layers of the neural networks are trained to automatically find suitable features from the training data.
Notes – other classification places
Subgroups of group G06V 10/40 focus on specific kinds of feature extraction techniques. These include:
- Features which describe characteristics of the entire image or an entire object (group G06V 10/42);
Note: Global feature extraction techniques often involve domain transformations, such as frequency domain transformation. The global descriptors contain numerical data, such as vectors or matrices, but they can also represent the image or object in an abstract form as a string of symbols from a predetermined alphabet, which are integrated using a grammar (covered by group G06V 10/424).
- Graph structures having vertices and edges (e.g. directed attributed graphs or trees) are another way of representing patterns in images; the vertices of such graph structures represent qualitative or quantitative feature measurements; the edges represent relations between them (covered by group G06V 10/426);
- Local features (covered by group G06V 10/44) build representations of the local image content. Examples of local features include luminance values or colour characteristics, potentially from more than three colour channels, local edges, corners, gradients and texture. Edges can be extracted by convolutions with specially designed filter masks (e.g. Prewitt, Sobel) or by convolutions with a numerical filter, e.g. wavelet filters (Haar, Daubechies), or by difference of Gaussians, Laplacian of Gaussians, Gabor filters etc. Local features such as edges and corners, which can be extracted by applying a pre-defined image operator, are also referred to as low-level features to distinguish them from features such as objects or events, which are extracted using a machine learning algorithm;
- Higher-level features, obtained e.g. by detecting silhouettes of shapes and describing them e.g. using a chain code, by a Fourier expansion of the contour, by curvature scale-space analysis, or by sampling points along object boundaries and quantifying their relative locations;
- Algorithms for evaluating the saliency of local image regions; selecting salient points as key points (covered by group G06V 10/46);
- For the purpose of feature extraction, techniques for converting image or video data to a different parameter space, e.g. using a Hough transform for detecting linear structures in images, or performing a conversion from the spatial domain to the frequency domain or vice versa (group G06V 10/48);
- Techniques for combining individual low-level features into feature vectors by first calculating local statistics of low-level image features in a block of pixels and subsequently generating histograms or deriving other statistical measures in a local neighbourhood (group G06V 10/50);
- Multi-scale feature extraction algorithms for analysing image or video data at different resolutions; scale space analysis, e.g. wavelet decompositions (group G06V 10/52);
- Techniques for describing textures, such as convolution with Gabor wavelets, grey-level co-occurrence matrices, or edge histograms (group G06V 10/54);
- Descriptors which capture colour properties of the image, such as colour histograms, possibly after conversion to a suitable colour space (group G06V 10/56);
- Descriptors which are specially designed for more than three colour channels, in particular for hyperspectral images which contain sensor readings in a multitude of different wavelengths not limited to the visual spectrum (group G06V 10/58);
- Descriptors obtained by integrating information about the imaging conditions, such as the position, the orientation, and the spectral properties of light sources, diffuse or specular reflections at object surfaces etc. (group G06V 10/60);
- Temporal descriptors derived from object movements, e.g. optical flow (group G06V 10/62).
Examples
Quantifying local image properties, in particular the local gradient, using a local probe
Different types of features used for object recognition, e.g. contours, line segments, continuous lines.
Querverweise
Nichteinschränkende Querverweise in anwendungsorientierte Klassifikationsstellen
Recognition of scene and scene-specific elements
| G06V 20/00 |
Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
| G06V 30/00 |
Image or video recognition or understanding of human-related, animal-related or biometric patterns in image or video data
| G06V 40/00 |
Recognition of fingerprints or palmprints
| G06V 40/12 |
Recognition of vascular patterns
| G06V 40/14 |
Recognition of human faces, e.g. facial parts, sketches or expressions within images or video data
| G06V 40/16 |
Recognition of eye characteristics within image or video data, e.g. of the iris
| G06V 40/18 |
Informative Querverweise
Spectrometry, measurement of colour
| G01J 3/46 |
Image analysis using feature-based methods, in particular for determination of transform parameters for the alignment of images
| G06T 7/33 |
Image analysis for depth or shape recovery
| G06T 7/50 |
Image contour coding, e.g. using detection of edges
| G06T 9/20 |
Glossar
BoW
| bag of words, a model originally developed for natural language processing; when applied to images, it represents an image by a histogram of visual words, each visual word representing a specific part of the feature space.
|
edge edges
| region in the image, at which the image exhibits a strong luminance gradient.
|
GLCM
| grey-level co-occurrence matrix
|
HOG
| histogram of oriented gradients, a feature descriptor described by N Dalal and B Triggs
|
SIFT
| scale-invariant feature transform, a feature detection algorithm
|
SURF
| speeded up robust features, a feature descriptor
|