Transforming Multiclass to Multilabel: Advanced Approaches in Image Classification

Computer vision tasks often involve multiclass image classification, where a picture is labelled by a specified class. Sometimes an image has several objects or qualities, requiring a more detailed method. This study converts multiclass image classification into multilabel image classification, allowing several labels per image. ResNet50, VGG16, and VGG19 are used for multilabel classification process. To improve multilabel classification and small dataset CIFAR-10, ESRGAN, a perceptual-driven approach for single image super-resolution that is able to produce photorealistic images, is used and redefining CIFAR-10 labels are employed. Traditional approaches cannot properly capture an image's intricacy or variety. The observed accuracy scores for VGG16, VGG19 and ResNet50 are 93.95 %, 94.18 %, 79.42 % respectively. The results demonstrate multilabel classification's ability to recognise visual elements and attributes.