Research On Color Building Image Segmentation Algorithm Based On Improved K-means

Image segmentation is a key step in image processing.Image segmentation also plays an important role in tasks such as 3D reconstruction in the field of architecture.However,it is difficult for existing segmentation algorithms to have a good performance when facing diverse segmentation objects,and thus cannot meet the needs of users.Therefore,it is very necessary to propose an efficient color building image segmentation algorithm.The basis of this analysis is that the essence of image segmentation is to cluster pixels.Combine the specific segmentation method with the segmented object,the K-means algorithm is analyzed and studied.The K-means algorithm used in image segmentation has the characteristics of strong adaptability and high efficiency,but there is a dependence on user input parameters.Besides,pixels can only be classified according to the Euclidean distance between pixel coordinate positions,which cannot fully describe the distribution characteristics of the pixels,which can easily cause misclassification of pixels.Based on these issues,this paper proposes a color building image segmentation algorithm based on improved K-means.First,through preprocessing,the algorithm clustering objects are converted into pixel blocks,which effectively reduces the calculation amount of the algorithm.Then obtain a reasonable number of classifications and a representative initial center point through the color histogram of the image.Then analyze the characteristics of color building images,use multi-dimensional feature constraints to calculate the similarity between pixel blocks,avoiding misclassification of pixel blocks,and improving the image segmentation effect.The innovation of this article includes the following two parts:(1)Aiming at the problem that the color image segmentation algorithm based on K-means is sensitive to initial parameters,a K-means parameter adaptive algorithm based on color histogram is proposed.The algorithm first creates a color histogram of the image in the HSI color space,and scans the color histogram vertically and horizontally to obtain peak points with higher density and a certain distance apart,and classifies the number of peak points and corresponding K pixel blocks as categories number and initial center point.Due to the use of the overall color distribution characteristics of the image,a more reasonable classification number and initial center point are obtained.(2)A multi-dimensional feature similarity calculation algorithm is proposed to make it suitable for segmentation of color architectural images.The algorithm analyzes the characteristics of color building images,and proposes to calculate the similarity of pixel blocks in color,texture and spatial position features,and combine the three as the final similarity to divide pixel blocks.A variety of feature constraints can be used to fully describe the local and global distribution of pixel blocks in the image,which improve the classification accuracy of pixel blocks,and effectively divide buildings into meaningful areas.