Penerapan Metode Harmonic Mean Filter  Dalam Mereduksi Gaussian Noise Pada Citra Digital

Agusti, Dhea; Nababan, Adli Abdillah

doi:10.32672/jnkti.v5i3.4468

Cited by 1 publication

(1 citation statement)

References 9 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Data pixel dari poster yang ada nantinya akan diubah ukurannya menjadi 317 x 230 x 3. Pixel adalah elemen citra yang memiliki nilai yang menunjukkan intensitas warna [24]. Label genre yang sesuai akan di ubah menjadi multi-hot vectors, detail Gambar 3.…”

Section: Hasil Dan Pembahasanunclassified

Transfer Learning to Predict Genre Based on Anime Posters

Kaka Kamaludin,

Woro Isti Rahayu,

Helmi Setywan

2023

J. Tek. Inform. (JUTIF)

View full text Add to dashboard Cite

Anime is an animated film with a distinctive graphic design originating from Japan, which is widely favored by various groups. anime itself has a genre like a movie in general, but there is a slight difference from ordinary films, anime has additional genres that are not in ordinary films, such as the Ecchi, Mahou Shoujou, Seinen, Shounen, and Josei genres. Since those genres only exist in anime, this research is devoted to predicting those anime genres. The prediction will use posters from the anime itself, with the help of image processing, namely the Convolutional Neural Network method and Transfer Learning. Transfer Learning will be implanted as a comparison of the performance of the existing architecture with the architecture that will be created, whether the architecture is able to process the dataset properly. The dataset to be used is a dataset of posters and csv documents containing images and details of the anime, the dataset contains anime data from 1980 to 2021 and contains 11651 anime poster data which has different resolution sizes. The ResNet50 model has the highest accuracy rate of 48% with a loss rate of 36%, while InceptionV3 produces 35% accuracy with 69% loss. At the time of testing ResNet50 gave the smallest genre percentage value of CustomModel and InceptionV3, while CustomModel gave the highest genre value. In addition to the value, all modes also predicted the genre well. Especially InceptionV3 is able to predict the music genre, because the music genre has a very small number of datasets, and this music genre is difficult to predict by the ResNet50 and CustomModel models.

show abstract

Section: Hasil Dan Pembahasanunclassified