Thai Leang Sung | Chonbuk National University (original) (raw)

Papers by Thai Leang Sung

We propose Identical-pair Adversarial Networks (iPANs) to solve image-to-image translation proble... more We propose Identical-pair Adversarial Networks (iPANs) to solve image-to-image translation problems, such as aerial-to-map, edge-to-photo, de-raining, and night-to-daytime. Our iPANs rely mainly on the effectiveness of adversarial loss function and its network architectures. Our iPANs consist of two main networks, an image transformation network T and a discriminative network D. We use U-NET for the transformation network T and a perceptual similarity network, which has two streams of VGG16 that share the same weights for network D. Our proposed adversarial losses play a minimax game against each other based on a real identical-pair and a fake identical-pair distinguished by the discriminative network D; e.g. a discriminative network D considers two inputs as a real pair only when they are identical, otherwise a fake pair. Meanwhile, the transformation network T tries to persuade the discriminator network D that the fake pair is a real pair. We experimented on several problems of image-to-image translation and achieved results that are comparable to those of some existing approaches, such as pix2pix, and PAN.

Conference Presentations by Thai Leang Sung

KIPS, 2019

In this paper, we propose a verifiable image transformation networks to transform face sketch to ... more In this paper, we propose a verifiable image transformation networks to transform face sketch to photo and vice versa. Face sketch-photo is very popular in computer vision applications. It has been used in some specific official departments such as law enforcement and digital entertainment. There are several existing face sketch-photo synthesizing methods that use feed-forward convolution neural networks; however, it is hard to assure whether the results of the methods are well mapped by depending only on loss values or accuracy results alone. In our approach, we use two Resnet encoder-decoder networks as image transformation networks. One is for sketch-photo and another is for photo-sketch. They depend on each other to verify their output results during training. For example, using photo-sketch transformation networks to verify the photo result of sketch-photo by inputting the result to the photo-sketch transformation networks and find loss between the reversed transformed result with ground-truth sketch. Likely, we can verify the sketch result as well in a reverse way. Our networks contain two loss functions such as sketch-photo loss and photo-sketch loss for the basic transformation stages and the other two-loss functions such as sketch-photo verification loss and photo-sketch verification loss for the verification stages. Our experiment results on CUFS dataset achieve reasonable results compared with the state-of-the-art approaches.

ISITC, 2019

An assessment of rice-seeds germination is a process of measuring the quality of the seeds for th... more An assessment of rice-seeds germination is a process of measuring the quality of the seeds for the benefit of rice planting farms in Surin Province and the neighboring areas in Thailand. We need specialists or experts in the area of agriculture to evaluate the seeds germination by classifying them into normal and abnormal seeds which require a lot of times and hard work. In this paper, we present our dataset collection and use convolutional neural networks (CNNs) to classify normal and abnormal Jasmine rice-seeds after germination of 7 days. Our purpose is to use deep learning technique such as CNNs to do the work of evaluation or classification instead of specialists or experts. We collected 1,562 sample images of Jasmine rice seed germination and categorized them into two groups, normal and abnormal. We also collected an extra 76 images mixing abnormal and normal together for testing. We set 75% of our dataset as a training set and 25% as a validation set. We build CNNs of 6 hidden layers and in each layer consists of convolution-pooling-Relu modules. It is the binary network that results as 0 and 1 which represent normal and abnormal. Therefore, in the last layer, we use a sigmoid function to acquire our score. Our experimental results show that the effectiveness of using CNNs in our work is very high. We obtain an average accuracy of 99.57% and loss 0.01% on training and accuracy 96.43% and loss 0.48% on validation.

ISITC, 2018

In this paper, we describe how to align a pair of high resolution RGB images using geometric tran... more In this paper, we describe how to align a pair of high resolution RGB images using geometric transformations. Opencv-python has been used for the implementation. ORB feature detection and points matching functions are presented here and homograpy transformation matrix is estimated. By applying this transformation matrix to the image that we want to align to; and overlap it to the target image, we obtain a result of an aligned image using stitching method.

ISITC, 2018

Edge detection in depth image remains a challenge in computer vision. In this paper, we propose a... more Edge detection in depth image remains a challenge in computer vision. In this paper, we propose an enhancement of depth edge detection using bilateral filtering and morphological operations such as erosion and dilation. The edge detection is done based on Canny Edge detection principle. The results have shown that this method provided better results than the method without the enhancement.

We propose Identical-pair Adversarial Networks (iPANs) to solve image-to-image translation proble... more We propose Identical-pair Adversarial Networks (iPANs) to solve image-to-image translation problems, such as aerial-to-map, edge-to-photo, de-raining, and night-to-daytime. Our iPANs rely mainly on the effectiveness of adversarial loss function and its network architectures. Our iPANs consist of two main networks, an image transformation network T and a discriminative network D. We use U-NET for the transformation network T and a perceptual similarity network, which has two streams of VGG16 that share the same weights for network D. Our proposed adversarial losses play a minimax game against each other based on a real identical-pair and a fake identical-pair distinguished by the discriminative network D; e.g. a discriminative network D considers two inputs as a real pair only when they are identical, otherwise a fake pair. Meanwhile, the transformation network T tries to persuade the discriminator network D that the fake pair is a real pair. We experimented on several problems of image-to-image translation and achieved results that are comparable to those of some existing approaches, such as pix2pix, and PAN.

KIPS, 2019

In this paper, we propose a verifiable image transformation networks to transform face sketch to ... more In this paper, we propose a verifiable image transformation networks to transform face sketch to photo and vice versa. Face sketch-photo is very popular in computer vision applications. It has been used in some specific official departments such as law enforcement and digital entertainment. There are several existing face sketch-photo synthesizing methods that use feed-forward convolution neural networks; however, it is hard to assure whether the results of the methods are well mapped by depending only on loss values or accuracy results alone. In our approach, we use two Resnet encoder-decoder networks as image transformation networks. One is for sketch-photo and another is for photo-sketch. They depend on each other to verify their output results during training. For example, using photo-sketch transformation networks to verify the photo result of sketch-photo by inputting the result to the photo-sketch transformation networks and find loss between the reversed transformed result with ground-truth sketch. Likely, we can verify the sketch result as well in a reverse way. Our networks contain two loss functions such as sketch-photo loss and photo-sketch loss for the basic transformation stages and the other two-loss functions such as sketch-photo verification loss and photo-sketch verification loss for the verification stages. Our experiment results on CUFS dataset achieve reasonable results compared with the state-of-the-art approaches.

ISITC, 2019

An assessment of rice-seeds germination is a process of measuring the quality of the seeds for th... more An assessment of rice-seeds germination is a process of measuring the quality of the seeds for the benefit of rice planting farms in Surin Province and the neighboring areas in Thailand. We need specialists or experts in the area of agriculture to evaluate the seeds germination by classifying them into normal and abnormal seeds which require a lot of times and hard work. In this paper, we present our dataset collection and use convolutional neural networks (CNNs) to classify normal and abnormal Jasmine rice-seeds after germination of 7 days. Our purpose is to use deep learning technique such as CNNs to do the work of evaluation or classification instead of specialists or experts. We collected 1,562 sample images of Jasmine rice seed germination and categorized them into two groups, normal and abnormal. We also collected an extra 76 images mixing abnormal and normal together for testing. We set 75% of our dataset as a training set and 25% as a validation set. We build CNNs of 6 hidden layers and in each layer consists of convolution-pooling-Relu modules. It is the binary network that results as 0 and 1 which represent normal and abnormal. Therefore, in the last layer, we use a sigmoid function to acquire our score. Our experimental results show that the effectiveness of using CNNs in our work is very high. We obtain an average accuracy of 99.57% and loss 0.01% on training and accuracy 96.43% and loss 0.48% on validation.

ISITC, 2018

In this paper, we describe how to align a pair of high resolution RGB images using geometric tran... more In this paper, we describe how to align a pair of high resolution RGB images using geometric transformations. Opencv-python has been used for the implementation. ORB feature detection and points matching functions are presented here and homograpy transformation matrix is estimated. By applying this transformation matrix to the image that we want to align to; and overlap it to the target image, we obtain a result of an aligned image using stitching method.

ISITC, 2018

Edge detection in depth image remains a challenge in computer vision. In this paper, we propose a... more Edge detection in depth image remains a challenge in computer vision. In this paper, we propose an enhancement of depth edge detection using bilateral filtering and morphological operations such as erosion and dilation. The edge detection is done based on Canny Edge detection principle. The results have shown that this method provided better results than the method without the enhancement.