Comparative Study on Classification of Images in Compressed and Uncompressed Domains

Image Classification is the basis of Computer vision. Classification with image data finds a variety of applications in various fields. A comparative study of Classifying the images in the compressed and uncompressed domain is exploited in this research. A garbage classification dataset with 6 different categories is employed for this purpose. Naitve Convolutional neural network is utilized as the classifier. Compressed classification is achieved by compressing the images using discrete wavelet transform of three levels and by feeding the compressed data to the model for classification. The accuracy achieved by classification in the uncompressed domain is about 82.1% with a loss of 0.892. The accuracy achieved in the compressed domain by the three levels of the discrete wavelet transform is 81.7%,81.2%, and 74.6% with losses of 0.793, 0.891, and 1.071 respectively. There is minimal difference in accuracy achieved between uncompressed and compressed domains. Image compression by discrete wavelet transform overperforms the uncompressed domain by several factors like size, elapsed time and memory space.