Penilaian Estetika Foto Makanan Berbasis Jaringan Saraf Dan Representasi Graf Adaptif

Ridhana, Irfan (2026) Penilaian Estetika Foto Makanan Berbasis Jaringan Saraf Dan Representasi Graf Adaptif. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221214-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (12MB) | Request a copy

Abstract

Visualisasi makanan memegang peranan penting dalam industri kuliner, khususnya dalam menarik minat konsumen melalui media digital. Namun, penilaian estetika foto makanan secara otomatis masih menjadi tantangan karena dipengaruhi oleh berbagai faktor visual, seperti warna, komposisi, tata letak, dan kejelasan objek. Pendekatan berbasis Convolutional Neural Networks (CNN) standar memiliki keterbatasan dalam menangkap hubungan antarbagian gambar, sedangkan pendekatan graf berbasis grid dapat memotong objek makanan secara kaku. Oleh karena itu, penelitian ini mengusulkan penilaian estetika foto makanan berbasis jaringan saraf dan representasi graf adaptif menggunakan segmentasi superpixel Simple Linear Iterative Clustering (SLIC). Region hasil segmentasi digunakan sebagai node, kemudian direpresentasikan menggunakan fitur visual dari InceptionResNet-v2 dan HRNet, serta fitur Color Moments dan Layout. Struktur graf dibangun menggunakan Superpixel Adjacency, K-Nearest Neighbor (KNN), dan kombinasi keduanya. Model yang dievaluasi meliputi Graph Convolutional Network (GCN), Principal Neighbourhood Aggregation (PNA), Graph Attention Network (GAT), dan Graph Attention Network v2 (GATv2) pada Gourmet Photography Dataset (GPD). Hasil pengujian menunjukkan bahwa skenario terbaik diperoleh pada proporsi 90% data latih dan 10% data uji. Model GATv2 dengan fitur HRNet dan topologi Superpixel Adjacency menghasilkan kinerja terbaik dengan akurasi 0,9167, precision 0,9154, recall 0,9172, dan F1-score 0,9161. Hasil ini menunjukkan bahwa kombinasi SLIC, HRNet, dan GATv2 mampu meningkatkan kinerja penilaian estetika foto makanan.
=================================================================================================================================
Food visualization plays a crucial role in the culinary industry, particularly in attracting consumer interest through digital media. However, automatic aesthetic assessment of food photography remains a challenge as it is influenced by various visual factors, such as color, composition, layout, and object clarity. Standard Convolutional Neural Network (CNN) approaches have limitations in capturing the relationships between different parts of an image, whereas grid-based graph approaches can rigidly segment food objects. Therefore, this study proposes a neural network-based food photo aesthetic assessment with adaptive graph representation using Simple Linear Iterative Clustering (SLIC) superpixel segmentation. The segmented regions are used as nodes, which are then represented using visual features from InceptionResNet-v2 and HRNet, alongside Color Moments and Layout features. The graph structure is constructed using Superpixel Adjacency, K-Nearest Neighbor (KNN), and a combination of both. The evaluated models include Graph Convolutional Network (GCN), Principal Neighbourhood Aggregation (PNA), Graph Attention Network (GAT), and Graph Attention Network v2 (GATv2) on the Gourmet Photography Dataset (GPD). The experimental results show that the best scenario was achieved with a proportion of 90% training data and 10% testing data. The GATv2 model, using HRNet features and Superpixel Adjacency topology, yielded the best performance with an accuracy of 0.9167, precision of 0.9154, recall of 0.9172, and F1-score of 0.9161. These findings demonstrate that the combination of SLIC, HRNet, and GATv2 can effectively enhance the performance of food photo aesthetic assessment.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Estetika, Foto Makanan, GATv2, HRNet, PNA, SLIC, Aesthetics, Food Photography, GATv2, HRNet, PNA, SLIC
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Irfan Ridhana
Date Deposited:	24 Jul 2026 02:20
Last Modified:	24 Jul 2026 02:20
URI:	http://repository.its.ac.id/id/eprint/137203

Actions (login required)

View Item