Implementasi Color Quantization pada Kompresi Citra Digital dengan Menggunakan Model Clustering Berdasarkan Nilai Max Variance pada Ruang Warna RGB

<p class="Abstrak">Kompresi citra dapat dilakukan dengan menggunakan <em>color quantization</em> di mana dengan mengurangi jumlah warna yang terdapat pada citra maka akan dapat mengurangi jumlah bit yang digunakan untuk merepresentasikan warna – warna tersebut. Semakin rendah jumlah warna yang dikurangi dalam rangka mencapai rasio kompresi yang optimal berdampak pada terdegradasinya kualitas dari citra. Secara umum <em>color quantization</em> menggunakan model <em>clustering </em>dalam proses pembentukan <em>color palette</em> yang akan digunakan sebagai referensi pada saat kuantisasi. Penelitian ini menggunakan model <em>clustering</em> berdasarkan nilai <em>max variance</em> pada <em>channel</em> RGB secara terpisah. Proses <em>clustering</em> dilakukan dengan membelah populasi <em>cluster </em>sebelumnya menggunakan nilai <em>mean</em> dari <em>channel </em>RGB yang memiliki nilai <em>variance </em>tertinggi. <em>Color palette</em> kemudian dibentuk menggunakan <em>centroid</em> hasil dari proses <em>clustering</em>. Percobaan pada beberapa citra uji dengan format 32bpp yang kemudian dikompresi menggunakan kuantisasi warna pada format 8bpp dan 4bpp memberikan kualitas dan rasio kompresi yang cukup baik yang diukur menggunakan ukuran MSE, PSNR dan CR di mana nilai MSE yang diperoleh berkisar 3.87 sampai 6.3 pada kuantisasi 8bpp dan 13.39 sampai 19.62 pada kuantisasi 4bpp. Sedangkan rasio kompresi yang diperoleh adalah sebesar 1.44 sampai 2.09 pada kuantisasi 8bpp dan 2.87 sampai 4.23 pada kuantisasi 4bpp.</p><p class="Abstrak"> </p><p class="Abstrak"><em><strong>Abstract</strong></em></p><p class="Judul2"><em>Image compression can be done by using color quantization where by reducing the number of colors contained in the image it can reduce the number of bits used to represent the colors. The lower the number of colors reduced in order to achieve the optimal compression ratio has an impact on the quality of the image. In general, color quantization uses clustering models in the process of constructing color palette that will be used as a reference during quantization. This study uses a clustering model based on the max variance value on the RGB channel separately. The clustering process is done by dividing the previous cluster population using the mean value of the RGB channel which has the highest variance value. The color palette is then formed using centroids resulting from the clustering process. Experiments on some test images in 32bpp format which are then compressed using color quantization in 8bpp and 4bpp formats provide a fairly good quality and compression ratio </em><em>with</em><em> MSE, PSNR and CR</em><em> assessment where the MSE values obtained ranged from 3.87 to 6.3 at 8bpp quantization and 13.39 to 19.62 at 4bpp quantization. While the compression ratio obtained is 1.44 to 2.09 at 8bpp quantization and 2.87 to 4.23 at 4bpp quantization </em></p><p class="Abstrak"><em><strong><br /></strong></em></p>

An Empirical Study on the Artificial Intelligence-Aided Quantitative Design of Art Images

This paper presents an indepth analysis and research on the quantitative design of fine art images through artificial intelligence algorithms. A CycleGAN-based network model for automatic generation of sketches of fine art images is constructed to extract the edge and contour features of fine art images. The network uses 512 × 1024 high-resolution art images as input and Pitchman as a discriminator. To further enhance the sketch generation effect, a bilateral filtering algorithm is added to the generator model for noise reduction, and then a K -means algorithm is used for color quantization to solve the problem of cluttered lines in the generated sketches. The experimental results show that the network model can effectively realize the automatic generation of art image sketches and can retain the detailed part of the costume information well. A rendering platform is built to realize the application of art image generation algorithms and coloring algorithms. The platform integrates the functions of image preprocessing, sketch generation, and sketch coloring, demonstrates the results of the main research content of this paper, and finally increases the interest of the system through the rendering function of the art image grid, which further improves the practicality of the platform.

Skin Cancer Classification Through Quantized Color Features and Generative Adversarial Network

Early interpretation of skin cancer through computer-aided diagnosis (CAD) tools reduced the intricacy of the treatments as it can attain a 95% recovery rate. To frame up with computer-aided diagnosis system, scientists adopted various artificial intelligence (AI) designed to receive the best classifiers among these diverse features. This investigation covers traditional color-based texture, shape, and statistical features of melanoma skin lesion and contrasted with suggested methods and approaches. The quantized color feature set of 4992 traits were pre-processed before training the model. The experimental images have combined images of naevus (1500), melanoma (1000), and basal cell carcinoma (500). The proposed methods handled issues like class imbalanced with generative adversarial networks (GAN). The recommended color quantization method with synthetic data generation increased the accuracy of the popular machine learning models as it gives an accuracy of 97.08% in random forest. The proposed model preserves a decent accuracy with KNN, adaboost, and gradient boosting.

Pembentua Pola Desain Motif Karawo Gorontalo Menggunakan K-Means Color Quantization dan Structured Forest Edge Detecion

<p class="Abstrak">Sulaman Karawo merupakan kerajinan tangan berupa sulaman khas dari daerah Gorontalo. Motif sulaman diterapkan secara detail berdasarkan suatu pola desain tertentu. Pola desain digambarkan pada kertas dengan berbagai panduannya. Gambar yang diterapkan pada pola memiliki resolusi sangat rendah dan harus mempertahankan bentuknya. Penelitian ini mengembangkan metode pembentukan pola desain motif Karawo dari citra digital. Proses dilakukan dengan pengolahan awal menggunakan <em>k-means color quantization (KMCQ)</em> dan deteksi tepi <em>structured forest</em>. Proses selanjutnya melakukan pengurangan resolusi menggunakan metode <em>pixelation</em> dan <em>binarization</em>. Luaran dari algoritma menghasilkan 3 citra berbeda dengan ukuran yang sama, yaitu: citra tepi, citra biner, dan citra berwarna. Ketiga citra tersebut selanjutnya dilakukan proses pembentukan pola desain motif Karawo dengan berbagai petunjuk pola bagi pengrajin. Hasil menunjukkan bahwa pola desain motif dapat digunakan dan dimengerti oleh para pengrajin dalam menerapkannya di sulaman Karawo. Pengujian nilai-nilai parameter dilakukan pada metode <em>k-means</em>, <em>gaussian filter</em>, <em>pixelation</em>, dan <em>binarization.</em> Parameter-parameter tersebut yaitu: k pada <em>k-means</em>, <em>kernel</em> pada <em>gaussian filter</em>, lebar piksel pada <em>pixelation</em>, dan nilai <em>threshold</em> pada <em>binarization</em>. Pengujian menunjukkan nilai terendah tiap parameter adalah k=4, kernel=3x3, lebar piksel=70, dan <em>threshold</em>=20. Hasil memperlihatkan makin tinggi nilai-nilai tersebut maka semakin baik pola desain motif yang dihasilkan. Nilai-nilai tersebut merupakan nilai parameter terendah dalam pembentukan pola desain motif berkualitas baik berdasarkan indikator-indikator dari desainer.</p><p class="Abstrak"> </p><p class="Abstrak"><em><strong>Abstract</strong></em></p><p class="Abstract"><em>Karawo embroidery is a unique handicraft from Gorontalo. The embroidery motif is applied in detail based on a certain design pattern. These patterns are depicted on paper with various guides. The image applied to the pattern is very low resolution and retains its shape. This study develops a method to generate a Karawo design pattern from a digital image. The process begins by using k-means color quantization (KMCQ) to reduce the number of colors and edge detection of the structured forest. The next process is to change the resolution using pixelation and binarization methods. The output algorithm produces 3 different state images of the same size, which are: edge image, binary image, and color image. These images are used in the formation of the Karawo motif design pattern. The motif contains various pattern instructions for the craftsman. The results show that it can be used and understood by the craftsmen in its application in Karawo embroidery. Testing parameter values on the k-means method, Gaussian filter, pixelation, and binarization. These parameters are k on KMCQ, the kernel on a gaussian filter, pixel width in pixelation, and threshold value in binarization. The results show that the lowest value of each parameter is k=4, kernel=3x3, pixel width=70, and threshold=20. The results show that the higher these values, the better the results of the pattern design motif. Those values are the lower input to generate a good quality pattern design based on the designer’s indicators.</em></p><p class="Abstrak"><em><strong><br /></strong></em></p>

Classification of color features in butterflies using the Support Vector Machine (SVM)

Research in digital images is expanding widely and includes several sectors. One sector currently being carried out research is in insects; specifically, butterflies are used as a dataset. A total of 890 types of butterflies divided into ten classes were used as a dataset and classified based on color. Ten types of butterflies include Danaus plexippus, Heliconius charitonius, Heliconius erato, Junonia coenia, Lycaena phlaeas, Nymphalis antiopa, Papilio cresphontes, Pieris rapae, Vanessa atalanta, Vanessa cardui. The process of extracting color features on butterfly wings uses the RGB method to become HSV color space with color quantization (CQ). The purpose of adding CQ is that the computation process is carried out faster without reducing the image's information. In the color feature extraction process, the image is converted into 3-pixel sizes and normalized. The process of normalizing the dataset has the aim that the value ranges in the dataset have the same value. The 890 butterfly dataset was classified using the Support Vector Machine (SVM) method. Based on this research process, the accuracy of the 256x160 pixel size is 72%, the 420x315 pixel is 75%, and the 768x576 pixel is 75%. The test results on a system with a 768x576 pixel get the highest results with a precision value of 74.6%, a recall of 72%, and an f-measure of 73.2% Keywords—image processing; classification; butterflies; color features; features extraction