You get a bonus - 1 coin for daily activity. Now you have 1 coin

5.5. MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES

Lecture



Discrete image represented by an array   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES can be fully described using the joint probability density of its elements. When specified in matrix form, this density is written as

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , (5.5.1а)

and in vector form - as

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , (5.5.1b)

Where   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES determines the order of joint density. If all elements of the image are statistically independent, then the joint probability density is equal to the product of one-dimensional unconditional densities.

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES . (5.5.2)

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES

Fig. 5.4.2. Energy spectra of Markov image-modeling processes (   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , on the vertical axis, the logarithmic scale): a is a separable spectrum; b - spectrum with circular symmetry.

The most common type of joint probability density is gaussian density.

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , (5.5.3)

Where   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES - vector covariance matrix   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES ,   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES - average value   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES and the symbol   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES marked matrix determinant   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES . The Gaussian density is a useful model of the joint probability density of the coefficients obtained as a result of unitary transformations of images. However, the Gaussian density is not suitable for describing the brightness of image elements, since the brightness can only be positive, and Gaussian random variables take both positive and negative values.

Expressions for joint densities that are not Gaussian are rarely found in the literature. Huns [13] developed a method for forming such high order densities based on a given unconditional first-order density and a given covariance matrix of ensemble elements. For zero-mean density, this procedure reduces to a linear transformation of a set of independent random variables.   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES whose joint probability density is

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES (5.5.4)

can be written as a product of given first-order densities. Then the desired joint probability density is

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , (5.5.5)

Where

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , (5.5.6)

but   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES denotes the determinant of the matrix   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES . Matrix columns   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES are eigenvectors of a given covariance matrix   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES and matrix   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES - diagonal and consists of the eigenvalues ​​of this matrix, and the relation

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES . (5.5.7)

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES

Fig. 5.5.1. Two-dimensional probability density pairs of correlated random variables   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES : a - probability density distribution of Laplace; b - Rayleigh probability distribution density.

In fig. 5.5.1 shows two-dimensional probability densities of a pair of correlated random variables, the unconditional distributions of which are Rayleigh or Laplace distributions. The multidimensional model with the Rayleigh distribution is useful in considering the joint probability density of the brightness of the image elements, and the model with the Laplace distribution is used to statistically describe the sequence of difference signals generated in the image coding systems using the prediction method.

The next chapter discusses the quantization technique, that is, the representation of samples using a discrete set of numbers, called quantization levels. Let be   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES denotes   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES level of quantization for the image element that occupies   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES -e place in the vector image   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES . Then the probability of obtaining one of the possible values ​​of the vector   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES can be expressed through the joint distribution of probabilities of sample values ​​as follows:

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , (5.5.8)

Where   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES . Usually for all components of the vector choose the same set of quantization levels and the joint probability distribution takes the form

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES . (5.5.9)

The probability distributions of sample values ​​can be estimated by measuring the corresponding frequencies. So, one-dimensional distribution   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES vector components

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES (5.5.10)

can be estimated by analyzing a large set of images belonging to the same class, such as fluorograms, aerial photographs of fields, etc. The estimate of the one-dimensional probability distribution is the distribution of relative frequencies

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , (5.5.11)

Where   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES - the total number of images studied, and   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES - number of shots for which   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES ,   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES . If the source of images is stationary, then one-dimensional distributions (5.5.10) will be the same for all components of the vector, i.e. they will not depend on   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES . In addition, if the source of images is ergodic, then ensemble averaging (measurements using a set of images) can be replaced by averaging over spatial coordinates. If the ergodicity assumption is true, then the one-dimensional distribution can be estimated from the frequencies

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , (5.5.12)

Where   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES - the number of elements of the investigated image, for which   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , and   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , but   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES .

In fig. 5.5.2 shows one-dimensional histograms of red, green and blue coordinates of the color of the color image "Portrait". In most natural images, dark elements are much larger than light ones, and the frequencies decrease with an increase in brightness approximately exponentially.

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES

Fig. 5.5.2. Typical histograms of red, green and blue color image color coordinates.

An estimate of the two-dimensional probability distribution for an ergodic source of images can be found using a second-order frequency distribution, which is obtained by counting the cases of the occurrence of certain pairs of element values ​​separated by a given distance. Assume that   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES and   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES there are two image elements separated by a length   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES which is tilted to the horizontal axis at an angle   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES (fig. 5.5.3). Since the elements form a rectangular lattice, the parameters   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES ,   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES may take only some discrete values. The relative frequency of the second order is

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES , (5.5.13)

Where   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES - the number of pairs of picture elements for which   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES and   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES . Denominator of a fraction (5.5.13)   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES represents the total number of pairs of elements that are separated by segments with the same parameters   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES . Due to edge effects   5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES .

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES

Fig. 5.5.3. Mutual arrangement of a pair of image elements.

  5.5.  MODELS OF DENSITY OF PROBABILITY OF DISCRETE IMAGES

Fig. 5.5.4. Second-order histograms for the “Portrait” image.

The second-order histograms for the “Portrait” image are presented in Fig.5.5.4. The elements of the image with the increase in the distance between them become less correlated, and the frequencies are distributed on the plane more evenly.


Comments


To leave a comment
If you have any suggestion, idea, thanks or comment, feel free to write. We really value feedback and are glad to hear your opinion.
To reply

Digital image processing

Terms: Digital image processing