Gray Image Quantization Method Based on New Threshold Optimizing Technique

Images are one of the most important media information in our world it used as inputs to many functions like classification, content-based image retrieval, image feature extraction, and many images processing purposes. In these fields, the optimal image quantization and reduction of gray levels methods are considered a significant preprocess that increases the efficiency and performance of the applications and enhances the precision of its results. Image quantization has a big effect on the use of huge image datasets used in these applications whereas it reduces the size of images by retaining features and the general structure of the image. This paper proposes a new image quantization method depend on optimal selection for threshold values depending on the PSNR metric. This method uses one of the Greedy programming algorithms that is the Divide and Conquer strategy. In this strategy, the selection of the threshold value that represents an optimum division position in the image gray levels range depends on the heuristic function. The heuristic function decides whichever threshold value (new gray level) from the image gray levels range is an optimal PSNR if it is selected and added to the quantized image compared with the original image. The experimental results show that the proposed method enhances the quantized image quality compared with standard quantization methods as standard uniform quantization and k-mean clustering method and other methods. The proposed method gets efficient results in PSNR and SSIM quality measures with increasing average ranged in [1.2%-5.5%] and [2%-6.5%], respectively, compared with other methods considering different gray levels and different image types. The method time complexity analysis is explained. The experimental results show the time consuming of applying the proposed method on many images and two big datasets.