Classification-oriented Image Quality Assessment Methods

With the rapid development of remote sensing technology, a large amount of remotesensing data has been provided for a variety of remote sensing applications. While due tothe lack of effective quality evaluation system and methods of remote sensing data inspecific application areas, there is great blindness in the acquisition, processing andapplication of remote sensing data. This lack of awareness of data availability will nodoubt affect research in image processing and analysis algorithms, and the evaluation ofvarious processing and analysis of image will be lack of effectiveness. All of these willhinder the expansion of the amplitude and deepening of the level of remote sensingapplications.This paper proposes two classification-oriented image quality assessment methods onthe basis of analysis of the quality of image classification. The first image qualityassessment method is based on image quality factors. For the first method, this paper firstanalyzes the impact of image quality factors on image quality, and then explores theconnection between image quality factors and image classification accuracy; the secondimage quality assessment method is based on the distribution of image features. For thefirst method, this paper first builds the image quality model which reflects the relationshipbetween the distribution of image features and image classification accuracy, and thenevaluates the image quality using the image quality evaluation model. Since the EMalgorithm can easily get trapped in local optimal solution, which may deviate far from thetrue value, this paper proposes a MCEM (Mean-Constrained Expectation Maximization)algorithm. Compared with the EM algorithm, both theory and experiments show thatMCEM algorithm can effectively improve the accuracy of parameter estimation.The experimental results using both the experimental simulation data and remotesensing data, verify the feasibility of the first method proposed in this paper. Theexperimental results using remote sensing data verify the validation of the second method.Experimental results show that the proposed second method of image quality evaluation,which is based on the distribution of image features can effectively estimate the accuracyof image classification.