Grzegorz Mikolajczak

Generation of a FIR filter by means of a neural network for improvement of the digital images obtained using the acquisition equipment based on the low quality CCD structure

The paper features the method of application of a neural network for improving the quality of the digital images generated by means of devices for backup and processing of data into a digital form which construction is based on the Charge Coupled Device (CCD) structure. In order to introduce the problem, the digital images were generated by means of two scanners (including a high class and a low class scanner) and afterwards the images were subject to an objective and a subjective evaluation. An objective evaluation was performed using two quality criteria, i.e. MSE (Mean Square Error) and Universal Image Quality Index. A FIR (Finite Impulse Response) filter applied for filtration of a low quality image was obtained as the result of the neural network learning process. The image so generated as the result of filtration features a superior quality in comparison to the original.

The Synchronization of the Images Based on Normalized Mean Square Error Algorithm

As it is known, to transform an analogue image into digital form it is necessary to undergo the processes of sampling and quantification. The first of them consists of downloading at defined intervals data from analogue image the second one approximates analogue levels of brightness due to the closest digital levels. Both processes are the reason of an errors formation. Those errors have significant influence on the fields of the digital images transformation, in which it is necessary to synchronize images. This problem becomes particularly significant when we use the images gained from two different sources (scanner, digital camera). Anyone who uses the terms concern to images’ transformation, knows that bad synchronization can lead to wrong results. In his article authors present the algorithm which eliminates problem of bad images adjustment. The paper features the method of determination of the rotation angle and axis based on computation of the Normalized Mean Square Error (NMSE) coefficient.

Generation of FIR filters by using neural networks to improve digital images

The aim of the article is to show correction possibilities of digital images, achieved by image acquisition tools built on low class CCD matrices, such as their quality were close to images achieved by high class tools. For this purpose the authors used the linear filter with 3×3 mask, which were generated with neural network.

Detection of noise in digital images by using the averaging filter name COV

One of the significant problems in digital signal processing is the filtering and reduction of undesired interference. Due to the abundance of methods and algorithms for processing signals characterized by complexity and effectiveness of removing noise from a signal, depending on the character and level of noise, it is difficult to choose the most effective method. So long as there is specific knowledge or grounds for certain assumptions as to the nature and form of the noise, it is possible to select the appropriate filtering method so as to ensure optimum quality. This chapter describes several methods for estimating the level of noise and presents a new method based on the properties of the smoothing filter.

Using a neural network to generate a FIR filter to improves digital images using a discrete convolution operation

The aim of the article is to show correction possibilities of digital images, achieved by image acquisition tools built on low class CCD matrices, such as their quality were close to images achieved by high class tools. For this purpose the authors used the linear filter with 3x3 mask, which were generated with neural network. Digital images were compared using the quality metrics such as MSE, NMSE and Q.

Using a Single-Layer Neural Network to Generate a FIR Filter That Improves Digital Images Using a Convolution Operation

The paper features the method of application of a neural network for improving the quality of the digital images generated by means of devices for backup and processing of data into a digital form which construction is based on the Charge Coupled Device (CCD) structure. In order to introduce the problem, the digital images were generated by means of two scanners (including a high class and a low class scanner) and afterwards the images were subject to an objective and a subjective evaluation. An objective evaluation was performed using two quality criteria, i.e. MSE (Mean Square Error) and NMSE (Normalized Mean Square Error). A FIR (Finite Impulse Response) filter applied for filtration of a low quality image was obtained as the result of the neural network learning process. The image so generated as the result of filtration features a superior quality in comparison to the original.

Estimation of the variance of noise in digital images using a median filter

This paper proposes a method for estimating noise based on the assumption that in the process of smoothing with median filter only noise damping takes place. Knowledge of noise damping coefficient and the selection of the appropriate area allows for more accurate estimation of the variance of the normal noise distribution.

Using the Q measure to create panoramic photographs

The authors presented a new method of creating panoramic photographs. A panoramic photograph is composed of a few images taken with a camera. The authors developed and tested an algorithm to create panoramic photographs which is based on a popular index applied in evaluation of quality of digital images using a reference image - the Universal Quality Image Index. The result of applying the suggested algorithm is an effective match of the sequence of partial digital photographs which make the panoramic photograph.

Differential approximation of the 2-D Laplace operator for edge detection in digital images

The paper presents the application of finite difference edge detection in digital images. For edge detection we use several methods. One of them is a method based on Laplace operator. This paper presents a differential approximation of the two-dimensional Laplace operator. The paper proposes a differential approximation, Laplace operator, based on 9-th lattice mask. Coefficients were determined using the Z transform. Optimization is based on the criterion of maximal compatibility differential approximation of Laplace filter with ideal. Mask parameters were chosen based on the analysis of the error function. Activity obtained filter has been tested on a digital image that contains many elements of geometry.

Estimation of the Level of Disturbance in Time Series Using a Median Filter

Information about the level of signal interference, allows you to select the appropriate method pre-processing information. Assuming that the disturbance is a process additive, a normal distribution can do this using the smoothing filters, and in particular the median filter. This chapter presents a method of estimating the level of disturbance, based on median filtration and the assumption that the smoothing process applies to noise, exclusively. The knowledge of a noise reduction coefficient enables the determining of an estimated quantity.