Download to read offline
More Related Content
Similar to Digital_Image_Classification.pptx
Digital_Image_Classification.pptx
- 1. Digital Image Classification BinodHumagain Instructor, LMTC
- 2. Contents Introduction SpectralSignature Training of Classifiers Feature Selection and Separability Analysis Image Classification Techniques Unsupervised Classification Supervised Classification Object Based Classification(OBIA) Accuracy Assessment
- 3.
- 4. Image Classification Groupingof Similar features Separation of dissimilar one Assigning class level to pixel Resulting in manageable size of class
- 6. Image Classification Digitalimage classification uses the spectral information represented by the digital numbers in one or more spectral bands, classify each individual pixel based on this spectral information. multispectral data are used to perform the classification, and the spectral pattern present within the data for each pixel is used as numerical basis for categorization.
- 7. Advantage of ImageClassification To analyze thematic characteristics of object based on brightness values in image To translate continuous variability of image data into map patterns that provide meaning to the user To obtain insight in the data with respect to ground cover and surface characteristics To find anomalous patterns in the image data set
- 8. Disadvantage of Imageclassification No use of other characteristics such as location, orientation, pattern, texture . . . Exist spectral overlap i.e. heterogeneous classes, mixed pixels (boundaries),… No categories in land use and land cover
- 9. Need of ImageClassification Cost efficient in the analyses of large data sets Results can be reproduced More objective then visual interpretation Effective analysis of complex multi-band (spectral) interrelationships Classification achieves data size reduction together with manual digitizing
- 10.
- 11. Spectral Signature The setof discrete spectral radiance measurements provided by the broad spectral bands of the sensor define the spectral signature of each class
- 12. Dimension of Data Spectral Dimensionality is determined by the number of sets of values being used in a process. In image processing, each band of data is a set of values. An image with four bands of data is said to be four-dimensional. In addition to raw bands, we can also use derived products like NDVI images and PCA Image
- 13. Measurement Vector Themeasurement vector of a pixel is the set of data file values for one pixel in all n bands. Although image data files are stored band-by band, it is often necessary to extract the measurement vectors for individual pixels. Generalized Measured Vector or Feature Vector
- 14. Mean Vector Whenthe measurement vectors of several pixels are analyzed, a mean vector is often calculated. This is the vector of the means of the data file values in each band. It has n elements. Generalized Mean Vector
- 15. Feature Space Afeature space image is a scatter plot of the pixel values of two bands of the imagery. The intensity of each portion of the feature space image is simply the number of pixels in the image that have that particular pair of x, y brightness values. The more intense the color, the higher the density of pixels with that particular combination of brightness's reds, yellows, and oranges indicate high densities the black, purple and blue, lower densities. A feature space image that compares uncorrelated bands such as infrared to visible usually gives a better feel for the partitioning than a feature space image comparing two correlated bands.
- 16. Feature Space Feature SpacePlot displaying high correlation between 2 bands of a satellite image Feature Space Plot displaying low correlation between 2 bands of a satellite image
- 17. Each featurevector is a point in the so-called feature space. A feature space image is simply a graph of the data file values of one band of data against the values of another band. N = the number of bands = dimensions …. an (n) dimensional data (feature) space. Features can be Raw bands Derived Images Feature Space
- 18. Similar objectsyields similar measurement results (feature vectors) i.e. nearby points in feature space correspond to similar objects Distance in feature space is related to dissimilarity Points that belong to the same class form a cloud in feature space Feature Space
- 20. Find anddetermine class a clusters Decide on decision boundaries Assign a class to each pixel Partition of Feature Space
- 21. Classification Types Commonclassification procedures can be divided into Supervised classification Unsupervised classification Object Based Image Classification(OBIA)
- 22. Supervised Classification Samplesof known identity are used to classify pixels of unknown identity Multivariate statistical parameters are calculated for the known samples. Every pixel is evaluated and assigned to the class which it most closely resemble digitally (in statistics). Hard Classification – a pixel is assigned to only one class. Called as supervised as the analyst supervises the algorithm by providing training sites
- 23. Unsupervised Classification Theidentities of land cover types (to be specified as classes) within a scene are generally not known a priori because either ground reference information is lacking or surface features within the scene are not well defined. The computer is required to group pixels with similar spectral characteristics into unique clusters according to some statistically determined criteria. Analyst then combine and re-labels the spectral clusters into information classes.
- 24. Object Based ImageClassification Object-Based Image Analysis (OBIA)- also called Geographic Object-Based Image Analysis (GEOBIA). OBIA is a sub-discipline of geoinformation science devoted to partitioning remote sensing imagery into meaningful image-objects, and assessing their characteristics through spatial, spectral and temporal scale. The fundamental step of any object-based image analysis is a segmentation of a scene-representing an image-into image objects.
- 25.
- 26. Supervised Classification: Overview The identity and location of some of the land cover types are known a priori through a combination of field work and experience. These locations or Samples or training sites or training areas are the homogeneous representative samples of the different surface cover types (information classes) of interest. The analyst supervises the algorithm by providing these training sites (Samples of known identity ) which are used to classify pixels of unknown identity Multivariate statistical parameters are calculated for these training sites. Every pixel is evaluated and assigned to the class which it most closely resemble digitally (in statistics). Hard Classification – a pixel is assigned to only one class.
- 27. Steps in SupervisedClassification
- 28. Training samples(also called samples) are sets of pixels that represent what is recognized as a Potential class. The system calculates statistics from the sample pixels to create a parametric signature for the class. It should be well distributed over the entire scene. Training Sample
- 31. Training datafor a class should be collected from homogeneous environment. Each site is usually composed of many pixels. Size: The general rule is that if training data is being collected from n bands then >10n pixels of training data is to be collected for each class This is sufficient to compute variance covariance matrices required by some classification algorithms. Selecting Training Sample
- 32. Remotely senseddata is highly correlated in nature Selection of variables showing strong discriminatory power and hence to avoid redundancies in feature set is termed as feature selection or feature extraction and is a key problem in classification Select a subset of features that are most discriminative. Select features which show less overlap between classes and hence are more discriminative. Feature Selection
- 33. Separability Analysis Thehighly correlated bands are rejected, and those with lesser or no correlation are selected for efficient analysis. Feature selection may involve both statistical and graphical analysis to determine the degree of between-class separability in the remote sensor training data. Using statistical methods, combinations of bands are normally ranked according to their potential ability to discriminate each class from all others using n bands at a time.
- 34. Selection of AppropriateAlgorithm Various supervised classification algorithms may be used to assign an unknown pixel to one of the classes. The choice of particular classifier depends on nature of input data and output required. Parametric: Parametric classification algorithms assume that the observed measurement vectors Xc , obtained for each class in each spectral band during the training phase are Gaussian in nature. Non Parametric classification algorithms make no such assumptions.
- 43.
- 44. Unsupervised classificationis the process where numerical operations are performed that search for natural groupings of the spectral properties of pixels, as examined in multispectral feature space. The clustering process results in a classification map consisting of m spectral classes. The analyst then attempts a posteriori (after the fact) to assign or transform the spectral classes into thematic information classes of interest (e.g., forest, agriculture). This may be difficult. Some spectral clusters may be meaningless because they represent mixed classes of Earth surface materials. The analyst must understand the spectral characteristics of the terrain well enough to be able to label certain clusters as specific information classes. Advantages of unsupervised classification is no extensive knowledge of the study area is required. Little user input is needed to perform unsupervised classification which minimizes the likelihood of human error. However, the analyst has little control of the classes generated and often these clusters contain multiple land covers making interpretation difficult Unsupervised Classification: Overview
- 45. Clustering parameters number of clusters size of cluster distance between the cluster centers cluster elimination value (threshold) Methods Unsupervised Slicing K-means clustering Clustering using the Iterative Self-Organizing Data Analysis Technique (ISODATA). Unsupervised Classification Technique
- 46.
- 48. K-Means Algorithm Iterativealgorithm Number of cluster K is known by user Most popular clustering algorithm Initialize randomly K cluster mean vectors Assign each pixel to any of the K clusters based on minimum feature distance After all pixels are assigned to K clusters, each cluster mean is recomputed Iterate till cluster mean vectors stabilize
- 49. ISODATA Algorithm IterativeSelf-Organizing Data Analysis Technique Generalization of K-Mean algorithm Consists of many user-specified parameters Minimum size of cluster Maximum size of cluster Maximum intra-cluster variance Minimum separation between pairs of clusters Maximum number of clusters Minimum number of clusters Maximum number of iteration
- 50. Unsupervised Classification Advantage, Disadvantages Advantages No prior knowledge is required Human Error is minimized Unique classes are recognized Disadvantages The output classes are spectral classes which we cannot exactly say as information classes and imposes constrain to our interpretation Additional Labelling is required Spectral properties vary over time, across image
- 51. Supervised vs. UnsupervisedClassification UNSUPERVISED APPROACH Considers only spectral distance measures Minimum user interaction Requires interpretation after classification Based on spectral groupings SUPERVISED APPROACH Incorporates prior knowledge Requires training set(samples) Based on spectral groupings More extensive user interaction
- 52. Classification: Object Based ImageAnalysis(OBIA)
- 53. Object Based ImageClassification Object-Based Image Analysis (OBIA)- also called Geographic Object-Based Image Analysis (GEOBIA). OBIA is a sub-discipline of geoinformation science devoted to partitioning remote sensing imagery into meaningful image-objects, and assessing their characteristics through spatial, spectral and temporal scale. The fundamental step of any object-based image analysis is a segmentation of a scene-representing an image-into image objects.
- 54. Object based imageclassification/Analysis Supervised and unsupervised classification is pixel-based; In other words, it creates square pixel and each pixel has a class; But object based classification groups pixels into representative vector shapes with size and geometry. Steps to perform object-based image analysis classification: Perform multiresolution segmentation Select training areas Define statistics Classify
- 55. Object based imageAnalysis Object-based image analysis (OBIA) segments an image by grouping pixels It doesn’t create single pixel. Instead, it generates objects with different geometries If you have the right image, objects can be so meaningful that it does the digitizing for you. For example, the segmentation results below highlights buildings The 2 most common segmentation algorithms are: Multiresolution segmentation in Ecognition Segment mean shift in ArcGIS
- 56. Object based imageAnalysis In Object-Based Image Analysis (OBIA) classification, you can use different methods to classify object. E.g. Can use SHAPE: if you want to classify buildings, you can use a shape statistic such as “rectangular fit”, this test an object’s geometry to the shape of a rectangle TEXTURE: Texture is the homogeneity of an object. For example, water is mostly homogeneous because it’s mostly dark blue. Bust forests have shadows and are mix of green and black SPECTRAL: You can use the mean value of spectral properties such as near-infrared, short-wave infrared, red, green or blue GEOGRAPHIC CONTENT: Objects have proximity and distance relationships between neighbors.
- 57. OBIA: Nearest NeighborClassification Nearest Neighbor(NN) classification is similar to supervised classification After multi-resolution segmentation, the user identifies sample sites for each land cover class. Next, they define statistics to classify image objects Finally, nearest neighbor classifies objects based on their resemblance to training sites and the statistics defined
- 58.
- 59. “A classification isnot complete until its accuracy has been assessed” Why? Curiosity: the desire to know how good something is. Crucial to know the quality of maps when using them to make resource decisions. Increase map quality by identifying and correcting sources of error. can compare the accuracy of various data, processing techniques, classification schemes, and interpreters. Accuracy Assessment
- 60. Comparison totwo sources of information Remote Sensing derived classification map Reference Test information The relationship between the two sets of information is expressed as a matrix known as Error matrix / confusion matrix /Contingency table Accuracy Assessment
- 61. Selection of SampleData: Sampling Scheme A combination of the following factors Class definition Design of sampling technique Determination of no. of samples for each class 61
- 62. Sampling Scheme? Questionsconcerning design of sample data: What are the map classes to be assessed and how are they distributed across the landscape? How many sample should be taken How should the samples be chosen Question concerning how the reference data should be collected What should be the source of reference data? How should reference data be collected? When should the reference data be collected? 62
- 63. Reference Data Issue 1:Choosing reference source Make sure you can actually extract from the reference source the information that you need for the classification scheme I.e. Aerial photos may not be good reference data if your classification scheme distinguishes four species of grass. You may need GPS’d ground data. 63
- 64. Reference data Issue 2:Determining size of reference plots Consider the extent and distribution of class i.e. take larger sample size for the features that cover larger area Consider heterogeneity in the scene i.e. increase sample size when the variation within the class is more 64
- 65. Sampling Methods Simple RandomSampling : observations are randomly placed. Stratified Random Sampling minimum number of observations are randomly placed in each category. 65
- 66. Sampling Methods Systematic Sampling Observationsare placed at equal intervals according to a strategy. Systematic Non-Aligned Sampling : a grid provides even distribution of randomly placed observations. 66
- 67. Sampling Methods Cluster Sampling Randomly placed“centroids” used as a base of several nearby observations. The nearby observations can be randomly selected, systematically selected, etc... 67
- 68.
- 69.
- 70.
- 71.
- 72.
- 73.
- 74.
- 75.
- 76.
- 77.
- 78.